<?xml version="1.0" encoding="UTF-8"?>
 <rdf:RDF xmlns="http://purl.org/rss/1.0/" xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:taxo="http://purl.org/rss/1.0/modules/taxonomy/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:cc="http://web.resource.org/cc/" xmlns:syn="http://purl.org/rss/1.0/modules/syndication/" xmlns:admin="http://webns.net/mvcb/">
  <channel rdf:about="http://pinboard.in">
    <title>Pinboard (arsyed)</title>
    <link>https://pinboard.in/u:arsyed/public/</link>
    <description>recent bookmarks from arsyed</description>
    <items>
      <rdf:Seq>	<rdf:li rdf:resource="https://arxiv.org/abs/2001.08361"/>
	<rdf:li rdf:resource="https://medium.com/@gajus/lessons-learned-scaling-postgresql-database-to-1-2bn-records-month-edc5449b3067"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1902.10176"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1811.01159"/>
	<rdf:li rdf:resource="https://medium.com/huggingface/training-larger-batches-practical-tips-on-1-gpu-multi-gpu-distributed-setups-ec88c3e51255"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1804.03235"/>
	<rdf:li rdf:resource="http://wesmckinney.com/blog/apache-arrow-pandas-internals/"/>
	<rdf:li rdf:resource="http://tomaugspurger.github.io/scalable-ml-02.html"/>
	<rdf:li rdf:resource="https://davidbarber.github.io/blog/2017/03/15/large-number-of-classes/"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1512.00567"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1305.6646"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1602.07714"/>
	<rdf:li rdf:resource="http://dl.acm.org/citation.cfm?id=2735474"/>
	<rdf:li rdf:resource="https://people.orie.cornell.edu/andrew/code/"/>
	<rdf:li rdf:resource="https://arxiv.org/abs/1606.04838"/>
	<rdf:li rdf:resource="http://ls2-www.cs.uni-dortmund.de/~schmidt/"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1606.00389"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1606.00399"/>
	<rdf:li rdf:resource="https://github.com/wiseio/paratext"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1605.09619"/>
	<rdf:li rdf:resource="http://www.jmlr.org/proceedings/papers/v51/yen16.pdf"/>
	<rdf:li rdf:resource="http://dl.acm.org/citation.cfm?id=2020420"/>
	<rdf:li rdf:resource="http://www.slideshare.net/SessionsEvents/alex-smola-professor-in-the-machine-learning-department-carnegie-mellon-university-at-mlconf-sf-111315"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1601.00393"/>
	<rdf:li rdf:resource="http://codinginparadise.org/ebooks/html/blog/nips_day_1.html"/>
	<rdf:li rdf:resource="http://matthewrocklin.com/blog/work/2015/08/28/Storage/"/>
	<rdf:li rdf:resource="http://dl.acm.org/citation.cfm?id=1626322"/>
	<rdf:li rdf:resource="http://www.unofficialgoogledatascience.com/2015/08/an-introduction-to-poisson-bootstrap_26.html?m=1"/>
	<rdf:li rdf:resource="http://stackoverflow.com/questions/25104733/how-to-efficiently-calculate-huge-matrix-multiplication-tfidf-features-in-pyth"/>
	<rdf:li rdf:resource="http://structureddata.org/"/>
	<rdf:li rdf:resource="http://bwlewis.github.io/GLM/"/>
	<rdf:li rdf:resource="http://papers.nips.cc/paper/5491-parallel-double-greedy-submodular-maximization"/>
	<rdf:li rdf:resource="http://machinelearning.wustl.edu/mlpapers/papers/icml2014c2_wei14"/>
	<rdf:li rdf:resource="http://jmlr.org/proceedings/papers/v37/leng15.html"/>
	<rdf:li rdf:resource="http://jmlr.org/proceedings/papers/v37/barbosa15.html"/>
	<rdf:li rdf:resource="http://lmwtree.devries.ninja/"/>
	<rdf:li rdf:resource="http://web.michaelchughes.com/"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1502.02606"/>
	<rdf:li rdf:resource="https://github.com/davidavdav/GaussianMixtures.jl"/>
	<rdf:li rdf:resource="http://people.csail.mit.edu/ludwigs/papers/icassp14_speakerlsh.pdf"/>
	<rdf:li rdf:resource="http://papers.nips.cc/paper/4362-fast-and-accurate-k-means-for-large-datasets"/>
	<rdf:li rdf:resource="http://www.clsp.jhu.edu/~delip/nocrawl/textgraphs09.pdf"/>
	<rdf:li rdf:resource="http://mitpress.mit.edu/books/large-scale-kernel-machines"/>
	<rdf:li rdf:resource="http://leon.bottou.org/publications/pdf/tricks-2012.pdf"/>
	<rdf:li rdf:resource="https://adventuresindatascience.wordpress.com/2014/12/30/minibatch-learning-for-large-scale-data-using-scikit-learn/"/>
	<rdf:li rdf:resource="http://jmlr.org/proceedings/papers/v32/gieseke14.html"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1411.4000"/>
	<rdf:li rdf:resource="http://static.googleusercontent.com/media/research.google.com/en/us/people/jeff/CIKM-keynote-Nov2014.pdf"/>
	<rdf:li rdf:resource="http://madalgo.au.dk/fileadmin/madalgo/Summer_School_2014/Introduction__2_.pdf"/>
	<rdf:li rdf:resource="http://jsatml.blogspot.com/2013/02/mini-batch-k-means-vs-elkan.html"/>
	<rdf:li rdf:resource="https://github.com/ptoulis/implicit-sgd"/>
	<rdf:li rdf:resource="http://curtis.ml.cmu.edu/w/courses/index.php/Machine_Learning_with_Large_Datasets_10-605_in_Spring_2014"/>
	<rdf:li rdf:resource="http://forty9ten.com/post/64676776664/avoiding-chef-suck-with-auto-scaling-groups"/>
	<rdf:li rdf:resource="http://dl.acm.org/citation.cfm?id=1135870"/>
	<rdf:li rdf:resource="https://etda.libraries.psu.edu/paper/9484/"/>
	<rdf:li rdf:resource="http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5206651&amp;abstractAccess=no&amp;userType=inst"/>
	<rdf:li rdf:resource="http://www.chioka.in/tea-time-with-convolutional-deep-belief-networks-for-scalable-unsupervised-learning-of-hierarchical-representations/"/>
	<rdf:li rdf:resource="https://bcb9f395-a-8ac90a7d-s-sites.googlegroups.com/a/vanhoucke.com/vincent/publications/vanhoucke-iclr14.pdf?attachauth=ANoY7coFEnIKISzOLVnX0s9zwv5-d15ocVtVDIqHbaIVBeqss0NoD9hk1FCvNXqNhjdi8LgmxxP-TqjEytBPFcWk91EG3f0WfvYYhfG1ftJq_8SmtTwVvwyn_aASMFY9ENoBoT5UlUv067EcBCi5_MHT2noyphZ9z_Oq0SJpLQOl8s8plVBpbgETF7ZkTf1GU2Mzll2AItk7GmOG9r8RhVYPTFN3mp35X4oML4yr3yFp4dVFA3-yKkw%3D&amp;attredirects=0"/>
	<rdf:li rdf:resource="http://www.cs.ucsb.edu/~veronika/MAE_LargeScaleClustering_strnadova.pdf"/>
	<rdf:li rdf:resource="http://www.mblondel.org/lightning/"/>
	<rdf:li rdf:resource="http://www.holovaty.com/writing/aws-notes/"/>
	<rdf:li rdf:resource="http://elelsee.com/"/>
	<rdf:li rdf:resource="http://www.jstatsoft.org/v55/i14"/>
	<rdf:li rdf:resource="https://code.facebook.com/posts/218678814984400/scaling-mercurial-at-facebook/"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1110.4198"/>
	<rdf:li rdf:resource="http://arxiv.org/abs/1112.5016"/>
	<rdf:li rdf:resource="http://stats.stackexchange.com/questions/63589/how-to-project-high-dimensional-space-into-a-two-dimensional-plane"/>
	<rdf:li rdf:resource="http://cseweb.ucsd.edu/~akmenon/ResearchExam.pdf"/>
	<rdf:li rdf:resource="http://news.ycombinator.com/item?id=5215884"/>
	<rdf:li rdf:resource="http://rapgenius.com/James-somers-herokus-ugly-secret-lyrics"/>
      </rdf:Seq>
    </items>
  </channel><item rdf:about="https://arxiv.org/abs/2001.08361">
    <title>[2001.08361] Scaling Laws for Neural Language Models</title>
    <dc:date>2020-03-05T20:32:07+00:00</dc:date>
    <link>https://arxiv.org/abs/2001.08361</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["We study empirical scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training, with some trends spanning more than seven orders of magnitude. Other architectural details such as network width or depth have minimal effects within a wide range. Simple equations govern the dependence of overfitting on model/dataset size and the dependence of training speed on model size. These relationships allow us to determine the optimal allocation of a fixed compute budget. Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence. "]]></description>
<dc:subject>neural-net error-landscape scaling</dc:subject>
<dc:identifier>https://pinboard.in/u:arsyed/b:74eddaf70e9a/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:error-landscape"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://medium.com/@gajus/lessons-learned-scaling-postgresql-database-to-1-2bn-records-month-edc5449b3067">
    <title>Lessons learned scaling PostgreSQL database to 1.2bn records/month</title>
    <dc:date>2019-11-22T16:31:29+00:00</dc:date>
    <link>https://medium.com/@gajus/lessons-learned-scaling-postgresql-database-to-1-2bn-records-month-edc5449b3067</link>
    <dc:creator>arsyed</dc:creator><dc:subject>database postgresql scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:6ab0c631ded3/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:database"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:postgresql"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1902.10176">
    <title>[1902.10176] A Memoization Framework for Scaling Submodular Optimization to Large Scale Problems</title>
    <dc:date>2019-03-26T21:31:35+00:00</dc:date>
    <link>https://arxiv.org/abs/1902.10176</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[We are motivated by large scale submodular optimization problems, where standard algorithms that treat the submodular functions in the \emph{value oracle model} do not scale. In this paper, we present a model called the \emph{precomputational complexity model}, along with a unifying memoization based framework, which looks at the specific form of the given submodular function. A key ingredient in this framework is the notion of a \emph{precomputed statistic}, which is maintained in the course of the algorithms. We show that we can easily integrate this idea into a large class of submodular optimization problems including constrained and unconstrained submodular maximization, minimization, difference of submodular optimization, optimization with submodular constraints and several other related optimization problems. Moreover, memoization can be integrated in both discrete and continuous relaxation flavors of algorithms for these problems. We demonstrate this idea for several commonly occurring submodular functions, and show how the precomputational model provides significant speedups compared to the value oracle model. Finally, we empirically demonstrate this for large scale machine learning problems of data subset selection and summarization.
]]></description>
<dc:subject>optimization submodularity scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:fb2ff3b19cd1/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1811.01159">
    <title>[1811.01159] Understanding and Comparing Scalable Gaussian Process Regression for Big Data</title>
    <dc:date>2018-11-06T03:58:20+00:00</dc:date>
    <link>https://arxiv.org/abs/1811.01159</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[ As a non-parametric Bayesian model which produces informative predictive distribution, Gaussian process (GP) has been widely used in various fields, like regression, classification and optimization. The cubic complexity of standard GP however leads to poor scalability, which poses challenges in the era of big data. Hence, various scalable GPs have been developed in the literature in order to improve the scalability while retaining desirable prediction accuracy. This paper devotes to investigating the methodological characteristics and performance of representative global and local scalable GPs including sparse approximations and local aggregations from four main perspectives: scalability, capability, controllability and robustness. The numerical experiments on two toy examples and five real-world datasets with up to 250K points offer the following findings. In terms of scalability, most of the scalable GPs own a time complexity that is linear to the training size. In terms of capability, the sparse approximations capture the long-term spatial correlations, the local aggregations capture the local patterns but suffer from over-fitting in some scenarios. In terms of controllability, we could improve the performance of sparse approximations by simply increasing the inducing size. But this is not the case for local aggregations. In terms of robustness, local aggregations are robust to various initializations of hyperparameters due to the local attention mechanism. Finally, we highlight that the proper hybrid of global and local scalable GPs may be a promising way to improve both the model capability and scalability for big data. ]]></description>
<dc:subject>gaussian-processes sparsity scaling big-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:4eed30178bed/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gaussian-processes"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sparsity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://medium.com/huggingface/training-larger-batches-practical-tips-on-1-gpu-multi-gpu-distributed-setups-ec88c3e51255">
    <title>💥 Training Neural Nets on Larger Batches: Practical Tips on 1-GPU, Multi-GPU &amp; Distributed setups</title>
    <dc:date>2018-10-15T15:37:27+00:00</dc:date>
    <link>https://medium.com/huggingface/training-larger-batches-practical-tips-on-1-gpu-multi-gpu-distributed-setups-ec88c3e51255</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["How can you train your model on large batches when your GPU can’t hold more than a few samples?

There are several tools, tips and tricks you can use to do that and I thought it would be nice to gather all the things I use and learned in a post."]]></description>
<dc:subject>pytorch scaling gpu parallel neural-net batch-size .* gradient-accumulation</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:84af6e932d9f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pytorch"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gpu"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:batch-size"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:.*"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gradient-accumulation"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1804.03235">
    <title>[1804.03235] Large scale distributed neural network training through online distillation</title>
    <dc:date>2018-04-11T04:20:02+00:00</dc:date>
    <link>https://arxiv.org/abs/1804.03235</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Techniques such as ensembling and distillation promise model quality improvements when paired with almost any base model. However, due to increased test-time cost (for ensembles) and increased complexity of the training pipeline (for distillation), these techniques are challenging to use in industrial settings. In this paper we explore a variant of distillation which is relatively straightforward to use as it does not require a complicated multi-stage setup or many new hyperparameters. Our first claim is that online distillation enables us to use extra parallelism to fit very large datasets about twice as fast. Crucially, we can still speed up training even after we have already reached the point at which additional parallelism provides no benefit for synchronous or asynchronous stochastic gradient descent. Two neural networks trained on disjoint subsets of the data can share knowledge by encouraging each model to agree with the predictions the other model would have made. These predictions can come from a stale version of the other model so they can be safely computed using weights that only rarely get transmitted. Our second claim is that online distillation is a cost-effective way to make the exact predictions of a model dramatically more reproducible. We support our claims using experiments on the Criteo Display Ad Challenge dataset, ImageNet, and the largest to-date dataset used for neural language modeling, containing ]]></description>
<dc:subject>neural-net teacher-student knowledge-distillation scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:acfa7c9359e5/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:teacher-student"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:knowledge-distillation"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://wesmckinney.com/blog/apache-arrow-pandas-internals/">
    <title>Apache Arrow and the &quot;10 Things I Hate About pandas&quot; - Wes McKinney</title>
    <dc:date>2017-10-05T13:06:46+00:00</dc:date>
    <link>http://wesmckinney.com/blog/apache-arrow-pandas-internals/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>apache-arrow pandas dataframe scaling interop</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:4c5d6b03b1de/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:apache-arrow"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pandas"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:dataframe"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:interop"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://tomaugspurger.github.io/scalable-ml-02.html">
    <title>datas-frame – Scalable Machine Learning (Part 2): Partial Fit</title>
    <dc:date>2017-09-25T23:24:11+00:00</dc:date>
    <link>http://tomaugspurger.github.io/scalable-ml-02.html</link>
    <dc:creator>arsyed</dc:creator><dc:subject>python pandas dask scaling sklearn out-of-core pipeline</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:b9d58ad8ec7d/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:python"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pandas"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:dask"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sklearn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:out-of-core"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pipeline"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://davidbarber.github.io/blog/2017/03/15/large-number-of-classes/">
    <title>Training with a large number of classes · David Barber</title>
    <dc:date>2017-06-15T17:46:52+00:00</dc:date>
    <link>https://davidbarber.github.io/blog/2017/03/15/large-number-of-classes/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>.tab machine-learning scaling classification</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:794d1870e36d/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:.tab"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:classification"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1512.00567">
    <title>[1512.00567] Rethinking the Inception Architecture for Computer Vision</title>
    <dc:date>2017-06-10T19:21:35+00:00</dc:date>
    <link>https://arxiv.org/abs/1512.00567</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Convolutional networks are at the core of most state-of-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we explore ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21.2% top-1 and 5.6% top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3.5% top-5 error on the validation set (3.6% error on the test set) and 17.3% top-1 error on the validation set.
]]></description>
<dc:subject>papers neural-net architecture inception computer-vision noise label-smoothing scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:d8e831b15453/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:architecture"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:inception"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:computer-vision"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:noise"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:label-smoothing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1305.6646">
    <title>[1305.6646] Normalized Online Learning</title>
    <dc:date>2016-12-20T04:10:14+00:00</dc:date>
    <link>https://arxiv.org/abs/1305.6646</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["We introduce online learning algorithms which are independent of feature scales, proving regret bounds dependent on the ratio of scales existent in the data rather than the absolute scale. This has several useful effects: there is no need to pre-normalize data, the test-time and test-space complexity are reduced, and the algorithms are more robust."]]></description>
<dc:subject>papers online-learning optimization scaling normalization</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:31a4b2da10e9/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:online-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:normalization"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1602.07714">
    <title>[1602.07714] Learning values across many orders of magnitude</title>
    <dc:date>2016-12-20T04:09:48+00:00</dc:date>
    <link>https://arxiv.org/abs/1602.07714</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Most learning algorithms are not invariant to the scale of the function that is being approximated. We propose to adaptively normalize the targets used in learning. This is useful in value-based reinforcement learning, where the magnitude of appropriate value approximations can change over time when we update the policy of behavior. Our main motivation is prior work on learning to play Atari games, where the rewards were all clipped to a predetermined range. This clipping facilitates learning across many different games with a single learning algorithm, but a clipped reward function can result in qualitatively different behavior. Using the adaptive normalization we can remove this domain-specific heuristic without diminishing overall performance.
]]></description>
<dc:subject>papers optimization normalization adaptive scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:2d9e0b6e55d3/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:normalization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:adaptive"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://dl.acm.org/citation.cfm?id=2735474">
    <title>Scaling up crowd-sourcing to very large datasets</title>
    <dc:date>2016-11-16T09:05:03+00:00</dc:date>
    <link>http://dl.acm.org/citation.cfm?id=2735474</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Crowd-sourcing has become a popular means of acquiring labeled data for many tasks where humans are more accurate than computers, such as image tagging, entity resolution, and sentiment analysis. However, due to the time and cost of human labor, solutions that rely solely on crowd-sourcing are often limited to small datasets (i.e., a few thousand items). This paper proposes algorithms for integrating machine learning into crowd-sourced databases in order to combine the accuracy of human labeling with the speed and cost-effectiveness of machine learning classifiers. By using active learning as our optimization strategy for labeling tasks in crowd-sourced databases, we can minimize the number of questions asked to the crowd, allowing crowd-sourced applications to scale (i.e., label much larger datasets at lower costs).

Designing active learning algorithms for a crowd-sourced database poses many practical challenges: such algorithms need to be generic, scalable, and easy to use, even for practitioners who are not machine learning experts. We draw on the theory of nonparametric bootstrap to design, to the best of our knowledge, the first active learning algorithms that meet all these requirements.

Our results, on 3 real-world datasets collected with Amazons Mechanical Turk, and on 15 UCI datasets, show that our methods on average ask 1--2 orders of magnitude fewer questions than the baseline, and 4.5--44× fewer than existing active learning algorithms.]]></description>
<dc:subject>papers active-learning crowdsourcing scaling big-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:5f0537fc2a5f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:active-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:crowdsourcing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://people.orie.cornell.edu/andrew/code/">
    <title>Scalable Kernel Learning and Gaussian Processes</title>
    <dc:date>2016-10-25T02:15:09+00:00</dc:date>
    <link>https://people.orie.cornell.edu/andrew/code/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>kernel-methods gaussian-processes kernel-learning scaling code</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:c92c533d082d/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:kernel-methods"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gaussian-processes"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:kernel-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:code"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://arxiv.org/abs/1606.04838">
    <title>[1606.04838] Optimization Methods for Large-Scale Machine Learning</title>
    <dc:date>2016-07-27T02:41:55+00:00</dc:date>
    <link>https://arxiv.org/abs/1606.04838</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["This paper provides a review and commentary on the past, present, and future of numerical optimization algorithms in the context of machine learning applications. Through case studies on text classification and the training of deep neural networks, we discuss how optimization problems arise in machine learning and what makes them challenging. A major theme of our study is that large-scale machine learning represents a distinctive setting in which the stochastic gradient (SG) method has traditionally played a central role while conventional gradient-based nonlinear optimization techniques typically falter. Based on this viewpoint, we present a comprehensive theory of a straightforward, yet versatile SG algorithm, discuss its practical behavior, and highlight opportunities for designing algorithms with improved performance. This leads to a discussion about the next generation of optimization methods for large-scale machine learning, including an investigation of two main streams of research on techniques that diminish noise in the stochastic directions and methods that make use of second-order derivative approximations."]]></description>
<dc:subject>papers optimization scaling machine-learning surveys</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:b5f348d419a0/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:surveys"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://ls2-www.cs.uni-dortmund.de/~schmidt/">
    <title>Melanie Schmidt - TU Dortmund, Informatik 2</title>
    <dc:date>2016-06-15T04:53:56+00:00</dc:date>
    <link>http://ls2-www.cs.uni-dortmund.de/~schmidt/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>people research clustering coreset scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:3e005f2014e5/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:people"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:research"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:clustering"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:coreset"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1606.00389">
    <title>[1606.00389] Stream Clipper: Scalable Submodular Maximization on Stream</title>
    <dc:date>2016-06-12T16:27:38+00:00</dc:date>
    <link>http://arxiv.org/abs/1606.00389</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Applying submodular maximization in the streaming setting is nontrivial because the commonly used greedy algorithm exceeds the fixed memory and computational limits typically needed during stream processing. We introduce a new algorithm, called stream clipper, that uses two thresholds to select elements either into a solution set S or an extra buffer B. The output is achieved by a greedy algorithm that starts from S and then, if needed, greedily adds elements from B. Swapping elements out of S may also be triggered lazily for further improvements, and elements may also be removed from B (and corresponding thresholds adjusted) in order to keep memory use bounded by a constant. Although the worst-case approximation factor does not outperform the previous worst-case of 1/2, stream clipper can perform better than 1/2 depending on the order of the elements in the stream. We develop the idea of an "order complexity" to characterize orders on which an approximation factor of 1−α can be achieved. In news and video summarization tasks, stream clipper significantly outperforms other streaming methods. It shows similar performance to the greedy algorithm but with less computation and memory costs.
]]></description>
<dc:subject>submodularity data-stream scaling optimization</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:77038c8a9d53/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:data-stream"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1606.00399">
    <title>[1606.00399] Scaling Submodular Maximization via Pruned Submodularity Graphs</title>
    <dc:date>2016-06-12T16:27:04+00:00</dc:date>
    <link>http://arxiv.org/abs/1606.00399</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[We propose a new random pruning method (called "submodular sparsification (SS)") to reduce the cost of submodular maximization. The pruning is applied via a "submodularity graph" over the n ground elements, where each directed edge is associated with a pairwise dependency defined by the submodular function. In each step, SS prunes a 1−1/c√ (for c>1) fraction of the nodes using weights on edges computed based on only a small number (O(logn)) of randomly sampled nodes. The algorithm requires logc√n steps with a small and highly parallelizable per-step computation. An accuracy-speed tradeoff parameter c, set as c=8, leads to a fast shrink rate 2‾√/4 and small iteration complexity log22√n. Analysis shows that w.h.p., the greedy algorithm on the pruned set of size O(log2n) can achieve a guarantee similar to that of processing the original dataset. In news and video summarization tasks, SS is able to substantially reduce both computational costs and memory usage, while maintaining (or even slightly exceeding) the quality of the original (and much more costly) greedy algorithm.
]]></description>
<dc:subject>submodularity optimization scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:efc127e33212/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://github.com/wiseio/paratext">
    <title>wiseio/paratext: A library for reading text files over multiple cores.</title>
    <dc:date>2016-06-08T20:33:59+00:00</dc:date>
    <link>https://github.com/wiseio/paratext</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["ParaText is a C++ library to read text files in parallel on multi-core machines. The alpha release includes a CSV reader and Python bindings."
]]></description>
<dc:subject>c++ libs text file parallel csv python pandas scaling big-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:20957188bee2/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:c++"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:libs"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:text"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:file"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:csv"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:python"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pandas"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1605.09619">
    <title>[1605.09619] Horizontally Scalable Submodular Maximization</title>
    <dc:date>2016-06-01T01:08:22+00:00</dc:date>
    <link>http://arxiv.org/abs/1605.09619</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["A variety of large-scale machine learning problems can be cast as instances of constrained submodular maximization. Existing approaches for distributed submodular maximization have a critical drawback: The capacity - number of instances that can fit in memory - must grow with the data set size. In practice, while one can provision many machines, the capacity of each machine is limited by physical constraints. We propose a truly scalable approach for distributed submodular maximization under fixed capacity. The proposed framework applies to a broad class of algorithms and constraints and provides theoretical guarantees on the approximation factor for any available capacity. We empirically evaluate the proposed algorithm on a variety of data sets and demonstrate that it achieves performance competitive with the centralized greedy solution."]]></description>
<dc:subject>submodularity scaling optimization</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:8634583f570a/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.jmlr.org/proceedings/papers/v51/yen16.pdf">
    <title>Scalable Exemplar Clustering and Facility Location viaAugmented Block Coordinate Descent with Column Generation</title>
    <dc:date>2016-05-13T19:20:52+00:00</dc:date>
    <link>http://www.jmlr.org/proceedings/papers/v51/yen16.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>scaling algorithms clustering facility-location submodularity optimization</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:b07d5d824e52/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:clustering"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:facility-location"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://dl.acm.org/citation.cfm?id=2020420">
    <title>Trading representability for scalability</title>
    <dc:date>2016-05-06T18:06:36+00:00</dc:date>
    <link>http://dl.acm.org/citation.cfm?id=2020420</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Support Vector Machines (SVMs) are among the most popular and successful classification algorithms. Kernel SVMs often reach state-of-the-art accuracies, but suffer from the curse of kernelization due to linear model growth with data size on noisy data. Linear SVMs have the ability to efficiently learn from truly large data, but they are applicable to a limited number of domains due to low representational power. To fill the representability and scalability gap between linear and nonlinear SVMs, we propose the Adaptive Multi-hyperplane Machine (AMM) algorithm that accomplishes fast training and prediction and has capability to solve nonlinear classification problems. AMM model consists of a set of hyperplanes (weights), each assigned to one of the multiple classes, and predicts based on the associated class of the weight that provides the largest prediction. The number of weights is automatically determined through an iterative algorithm based on the stochastic gradient descent algorithm which is guaranteed to converge to a local optimum. Since the generalization bound decreases with the number of weights, a weight pruning mechanism is proposed and analyzed. The experiments on several large data sets show that AMM is nearly as fast during training and prediction as the state-of-the-art linear SVM solver and that it can be orders of magnitude faster than kernel SVM. In accuracy, AMM is somewhere between linear and kernel SVMs. For example, on an OCR task with 8 million highly dimensional training examples, AMM trained in 300 seconds on a single-core processor had 0.54% error rate, which was significantly lower than 2.03% error rate of a linear SVM trained in the same time and comparable to 0.43% error rate of a kernel SVM trained in 2 days on 512 processors. The results indicate that AMM could be an attractive option when solving large-scale classification problems. The software is available at www.dabi.temple.edu/~vucetic/AMM.html.

]]></description>
<dc:subject>papers machine-learning scaling classification ensemble svm hyperplane sgd</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:bcf08db0fead/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:classification"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ensemble"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:svm"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:hyperplane"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sgd"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.slideshare.net/SessionsEvents/alex-smola-professor-in-the-machine-learning-department-carnegie-mellon-university-at-mlconf-sf-111315">
    <title>Fast, Cheap and Deep: Scaling machine learning (Alex Smola)</title>
    <dc:date>2016-03-09T22:33:25+00:00</dc:date>
    <link>http://www.slideshare.net/SessionsEvents/alex-smola-professor-in-the-machine-learning-department-carnegie-mellon-university-at-mlconf-sf-111315</link>
    <dc:creator>arsyed</dc:creator><dc:subject>slides machine-learning scaling tensorflow mxnet</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:f3b218f65ab4/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:slides"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:tensorflow"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mxnet"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1601.00393">
    <title>[1601.00393] On the Reducibility of Submodular Functions</title>
    <dc:date>2016-01-10T05:48:56+00:00</dc:date>
    <link>http://arxiv.org/abs/1601.00393</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["The scalability of submodular optimization methods is critical for their usability in practice. In this paper, we study the reducibility of submodular functions, a property that enables us to reduce the solution space of submodular optimization problems without performance loss. We introduce the concept of reducibility using marginal gains. Then we show that by adding perturbation, we can endow irreducible functions with reducibility, based on which we propose the perturbation-reduction optimization framework. Our theoretical analysis proves that given the perturbation scales, the reducibility gain could be computed, and the performance loss has additive upper bounds. We further conduct empirical studies and the results demonstrate that our proposed framework significantly accelerates existing optimization methods for irreducible submodular functions with a cost of only small performance losses."]]></description>
<dc:subject>papers submodularity optimization reducibility scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:8f6559b6b624/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:reducibility"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://codinginparadise.org/ebooks/html/blog/nips_day_1.html">
    <title>Brad Neuberg: NIPS Day 1: Tutorials on Scaling Deep Learning, Probabilistic Programming, and Reinforcement Learning</title>
    <dc:date>2015-12-10T14:45:58+00:00</dc:date>
    <link>http://codinginparadise.org/ebooks/html/blog/nips_day_1.html</link>
    <dc:creator>arsyed</dc:creator><dc:subject>nips-2015 jeff-dean google scaling machine-learning deep-learning swarch distcomp tensorflow reinforcement-learning</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:afe8cd65e418/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:nips-2015"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:jeff-dean"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:google"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:deep-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:swarch"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:distcomp"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:tensorflow"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:reinforcement-learning"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://matthewrocklin.com/blog/work/2015/08/28/Storage/">
    <title>Efficient Tabular Storage</title>
    <dc:date>2015-10-21T18:11:07+00:00</dc:date>
    <link>http://matthewrocklin.com/blog/work/2015/08/28/Storage/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["t turns out that these principles are actually quite easy to implement with the right tools (thank you #PyData) The rest of this post will talk about a tiny 500 line project, Castra, that implements these princples and gets good speedups on biggish Pandas data.

Castra

With these goals in mind we built Castra, a binary partitioned compressed columnstore with builtin support for categoricals and integration with both Pandas and dask.dataframe."]]></description>
<dc:subject>dask csv python disk memory storage scaling performance bandwidth big-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:4ea70e66f059/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:dask"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:csv"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:python"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:disk"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:memory"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:storage"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:performance"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:bandwidth"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://dl.acm.org/citation.cfm?id=1626322">
    <title>Scaling high-order character language models to gigabytes</title>
    <dc:date>2015-10-13T01:15:44+00:00</dc:date>
    <link>http://dl.acm.org/citation.cfm?id=1626322</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[We describe the implementation steps required to scale high-order character language models to gigabytes of training data without pruning. Our online models build character-level PAT trie structures on the fly using heavily data-unfolded implementations of an mutable daughter maps with a long integer count interface. Terminal nodes are shared. Character 8-gram training runs at 200,000 characters per second and allows online tuning of hyperparameters. Our compiled models precompute all probability estimates for observed n-grams and all interpolation parameters, along with suffix pointers to speedup context computations from proportional to n-gram length to a constant. The result is compiled models that are larger than the training models, but execute at 2 million characters per second on a desktop PC. Cross-entropy on held-out data shows these models to be state of the art in terms of performance.

]]></description>
<dc:subject>papers language-model ngram scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:e565721b2ff1/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:language-model"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ngram"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.unofficialgoogledatascience.com/2015/08/an-introduction-to-poisson-bootstrap_26.html?m=1">
    <title>The Unofficial Google Data Science Blog: An Introduction to the Poisson Bootstrap</title>
    <dc:date>2015-10-12T16:35:13+00:00</dc:date>
    <link>http://www.unofficialgoogledatascience.com/2015/08/an-introduction-to-poisson-bootstrap_26.html?m=1</link>
    <dc:creator>arsyed</dc:creator><dc:subject>statcomp bootstrap poisson scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:8d942d30e39e/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:statcomp"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:bootstrap"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:poisson"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://stackoverflow.com/questions/25104733/how-to-efficiently-calculate-huge-matrix-multiplication-tfidf-features-in-pyth">
    <title>numpy - How to efficiently calculate huge matrix multiplication (tfidf features) in Python? - Stack Overflow</title>
    <dc:date>2015-08-01T19:33:58+00:00</dc:date>
    <link>http://stackoverflow.com/questions/25104733/how-to-efficiently-calculate-huge-matrix-multiplication-tfidf-features-in-pyth</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["The formula from sklearn.random_projection.johnson_lindenstrauss_min_dim shows that to preserve up to a 10% tolerance, you only need johnson_lindenstrauss_min_dim(350363, .1) 10942 features. This is an upper bound, so you may be able to get away with much less. Even 1% tolerance would only need johnson_lindenstrauss_min_dim(350363, .01) 1028192 features which is still significantly less than you have right now."

"If the issue is that you cannot store the "final matrix" in memory either, I would recommend working with the data in an HDF5Store (as seen in pandas using pytables). This link has some good starter code, and you could iteratively calculate chunks of your dot product and write to disk. I have been using this extensively in a recent project on a 45GB dataset, and could provide more help if you decide to go this route."

"What you could do is slice a row and a column of X, multiply those and save the resulting row to a file. Then move to the next row and column."]]></description>
<dc:subject>scicomp machine-learning scaling performance tips random-projections distance pairwise</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:a029e070e909/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scicomp"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:performance"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:tips"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:random-projections"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:distance"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pairwise"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://structureddata.org/">
    <title>Structured Data | Thoughts on: Big Data, Hadoop, Databases, Platform, Performance &amp; Scalability</title>
    <dc:date>2015-07-15T09:53:04+00:00</dc:date>
    <link>http://structureddata.org/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>blogs database performance scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:492c6b777deb/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:blogs"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:database"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:performance"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://bwlewis.github.io/GLM/">
    <title>GLMs, abridged</title>
    <dc:date>2015-06-26T02:56:46+00:00</dc:date>
    <link>http://bwlewis.github.io/GLM/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["This note grew out of our own desire to better understand how to go about solving generalized linear models in practice. We highlight aspects of GLM implementations that we find particularly interesting. We present some reference implementations stripped down to illuminate core ideas; often with just a few lines of code. Finally, we discuss details that enable the development of effective distributed parallel implementations suitable for solution of large-scale problems."]]></description>
<dc:subject>statistics statcomp glm R optimization scaling big-data .read</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:bf5e9cf42624/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:statistics"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:statcomp"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:glm"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:R"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:.read"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://papers.nips.cc/paper/5491-parallel-double-greedy-submodular-maximization">
    <title>Parallel Double Greedy Submodular Maximization</title>
    <dc:date>2015-06-22T12:59:24+00:00</dc:date>
    <link>http://papers.nips.cc/paper/5491-parallel-double-greedy-submodular-maximization</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["Many machine learning problems can be reduced to the maximization of submodular functions. Although well understood in the serial setting, the parallel maximization of submodular functions remains an open area of research with recent results only addressing monotone functions. The optimal algorithm for maximizing the more general class of non-monotone submodular functions was introduced by Buchbinder et al. and follows a strongly serial double-greedy logic and program analysis. In this work, we propose two methods to parallelize the double-greedy algorithm. The first, coordination-free approach emphasizes speed at the cost of a weaker approximation guarantee. The second, concurrency control approach guarantees a tight 1/2-approximation, at the quantifiable cost of additional coordination and reduced parallelism. As a consequence we explore the trade off space between guaranteed performance and objective optimality. We implement and evaluate both algorithms on multi-core hardware and billion edge graphs, demonstrating both the scalability and tradeoffs of each approach."]]></description>
<dc:subject>papers algorithms maximization parallel scaling submodularity</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:fe9252d4346f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:maximization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://machinelearning.wustl.edu/mlpapers/papers/icml2014c2_wei14">
    <title>Fast Multi-stage Submodular Maximization</title>
    <dc:date>2015-06-22T12:52:57+00:00</dc:date>
    <link>http://machinelearning.wustl.edu/mlpapers/papers/icml2014c2_wei14</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Abstract: We introduce a new multi-stage algorithmic framework for submodular maximization. We are motivated by extremely large scale machine learning problems, where both storing the whole data for function evaluation and running the standard accelerated greedy algorithm are prohibitive. We propose a multi-stage framework (called MultGreed), where at each stage we apply an approximate greedy procedure to maximize surrogate submodular functions. The surrogates serve as proxies for a target submodular function but require less memory and are easy to evaluate. We theoretically analyze the performance guarantee of the multi-stage framework, and give examples on how to design instances of MultGreed for a broad range of natural submodular functions. We show that MultGreed performs very close to the standard greedy algorithm, given appropriate surrogate functions, and argue how our framework can easily be integrated with distributive algorithms for optimization. We complement our theory by empirically evaluating on several real world problems, including data subset selection on millions of speech samples, where MultGreed yields at least a thousand times speedup and superior results over the state-of-the-art selection methods.
]]></description>
<dc:subject>papers algorithms maximization scaling multi-stage submodularity</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:29f081fe7e66/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:maximization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:multi-stage"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://jmlr.org/proceedings/papers/v37/leng15.html">
    <title>Hashing for Distributed Data | ICML 2015 | JMLR W&amp;CP</title>
    <dc:date>2015-06-03T20:20:05+00:00</dc:date>
    <link>http://jmlr.org/proceedings/papers/v37/leng15.html</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Recently, hashing based approximate nearest neighbors search has attracted much attention. Extensive centralized hashing algorithms have been proposed and achieved promising performance. However, due to the large scale of many applications, the data is often stored or even collected in a distributed manner. Learning hash functions by aggregating all the data into a fusion center is infeasible because of the prohibitively expensive communication and computation overhead. In this paper, we develop a novel hashing model to learn hash functions in a distributed setting. We cast a centralized hashing model as a set of subproblems with consensus constraints. We find these subproblems can be analytically solved in parallel on the distributed compute nodes. Since no training data is transmitted across the nodes in the learning process, the communication cost of our model is independent to the data size. Extensive experiments on several large scale datasets containing up to 100 million samples demonstrate the efficacy of our method.
]]></description>
<dc:subject>papers algorithms hashing knn approximation parallel scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:e274080844c8/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:hashing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:knn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:approximation"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://jmlr.org/proceedings/papers/v37/barbosa15.html">
    <title>The Power of Randomization: Distributed Submodular Maximization on Massive Datasets | ICML 2015 | JMLR W&amp;CP</title>
    <dc:date>2015-06-03T20:16:46+00:00</dc:date>
    <link>http://jmlr.org/proceedings/papers/v37/barbosa15.html</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["A wide variety of problems in machine learning, including exemplar clustering, document summarization, and sensor placement, can be cast as constrained submodular maximization problems. Unfortunately, the resulting submodular optimization problems are often too large to be solved on a single machine. We consider a distributed, greedy algorithm that combines previous approaches with randomization. The result is an algorithm that is embarrassingly parallel and achieves provable, constant factor, worst-case approximation guarantees. In our experiments, we demonstrate its efficiency in large problems with different kinds of constraints with objective values always close to what is achievable in the centralized setting."]]></description>
<dc:subject>papers optimization maximization scaling parallel randomization submodularity</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:1b995fb7587b/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:optimization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:maximization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:randomization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://lmwtree.devries.ninja/">
    <title>LMW-tree by cmdevries</title>
    <dc:date>2015-06-01T20:08:11+00:00</dc:date>
    <link>http://lmwtree.devries.ninja/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[LMW-tree is a generic template library written in C++ that implements several algorithms that use the m-way nearest neighbor tree structure to store their data. The algorithms and data structures are generic to support different data representations and functionality.

]]></description>
<dc:subject>c++ libs clustering ir data-structures k-tree knn scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:7bc4144ab950/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:c++"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:libs"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:clustering"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ir"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:data-structures"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:k-tree"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:knn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://web.michaelchughes.com/">
    <title>Michael C. Hughes</title>
    <dc:date>2015-05-27T15:13:25+00:00</dc:date>
    <link>http://web.michaelchughes.com/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>people research machine-learning nonparametric scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:e44703c22a2e/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:people"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:research"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:nonparametric"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1502.02606">
    <title>[1502.02606] The Power of Randomization: Distributed Submodular Maximization on Massive Datasets</title>
    <dc:date>2015-05-16T15:00:30+00:00</dc:date>
    <link>http://arxiv.org/abs/1502.02606</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[A wide variety of problems in machine learning, including exemplar clustering, document summarization, and sensor placement, can be cast as constrained submodular maximization problems. Unfortunately, the resulting submodular optimization problems are often too large to be solved on a single machine. We develop a simple distributed algorithm that is embarrassingly parallel and it achieves provable, constant factor, worst-case approximation guarantees. In our experiments, we demonstrate its efficiency in large problems with different kinds of constraints with objective values always close to what is achievable in the centralized setting.
]]></description>
<dc:subject>papers scaling randomization submodularity</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:f1f0bdd32c85/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:randomization"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:submodularity"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://github.com/davidavdav/GaussianMixtures.jl">
    <title>davidavdav/GaussianMixtures.jl</title>
    <dc:date>2015-04-18T10:46:22+00:00</dc:date>
    <link>https://github.com/davidavdav/GaussianMixtures.jl</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["In training the parameters of a GMM using the Expectation Maximization (EM) algorithm, the inner loop (computing the Baum-Welch statistics) can be executed efficiently using Julia's standard parallelization infrastructure, e.g., by using SGE. We further support very large data (larger than will fit in the combined memory of the computing cluster) though BigData, which has now been incorporated in this package."]]></description>
<dc:subject>julia libs gmm scaling parallel</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:3ff5630568b6/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:julia"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:libs"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gmm"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://people.csail.mit.edu/ludwigs/papers/icassp14_speakerlsh.pdf">
    <title>LARGE-SCALE SPEAKER IDENTIFICATION</title>
    <dc:date>2015-02-17T20:14:50+00:00</dc:date>
    <link>http://people.csail.mit.edu/ludwigs/papers/icassp14_speakerlsh.pdf</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["Speaker identification is one of the main tasks in speech processing. In addition to identification accuracy, large-scale applications of speaker identification give rise to another challenge: fast search in the database of speakers. In this paper, we propose a system based on i-vectors, a current approach for speaker identification, and locality sensitive hashing, an algorithm for fast nearest neighbor search in high dimensions. The connection between the two techniques is the cosine distance: on the one hand, we use the cosine distance to compare i-vectors, on the other hand, locality sensitive hashing allows us to quickly approximate the cosine distance in our retrieval procedure. We evaluate our approach on a realistic data set from YouTube with about 1,000 speakers. The results show that our algorithm is approximately one to two orders of magnitude faster than a linear search while maintaining the identification accuracy of an i-vector-based system."]]></description>
<dc:subject>papers speaker-recognition i-vector scaling searching indexing knn hashing</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:816761248e4c/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:speaker-recognition"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:i-vector"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:searching"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:indexing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:knn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:hashing"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://papers.nips.cc/paper/4362-fast-and-accurate-k-means-for-large-datasets">
    <title>Fast and Accurate k-means For Large Datasets</title>
    <dc:date>2015-02-16T19:35:10+00:00</dc:date>
    <link>http://papers.nips.cc/paper/4362-fast-and-accurate-k-means-for-large-datasets</link>
    <dc:creator>arsyed</dc:creator><dc:subject>papers clustering k-means scaling coreset</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:a03394334f35/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:clustering"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:k-means"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:coreset"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.clsp.jhu.edu/~delip/nocrawl/textgraphs09.pdf">
    <title>Ranking and Semi-supervised Classification on Large Scale Graphs Using Map-Reduce</title>
    <dc:date>2015-02-16T18:35:22+00:00</dc:date>
    <link>http://www.clsp.jhu.edu/~delip/nocrawl/textgraphs09.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>papers ranking gbl graph mapReduce scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:75461509862f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ranking"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gbl"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:graph"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mapReduce"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://mitpress.mit.edu/books/large-scale-kernel-machines">
    <title>Large-Scale Kernel Machines | The MIT Press</title>
    <dc:date>2015-02-16T02:08:53+00:00</dc:date>
    <link>http://mitpress.mit.edu/books/large-scale-kernel-machines</link>
    <dc:creator>arsyed</dc:creator><dc:subject>books machine-learning kernel-methods scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:01ec3d0a9a25/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:books"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:kernel-methods"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://leon.bottou.org/publications/pdf/tricks-2012.pdf">
    <title>Stochastic Gradient Descent Tricks</title>
    <dc:date>2015-02-16T01:47:02+00:00</dc:date>
    <link>http://leon.bottou.org/publications/pdf/tricks-2012.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>machine-learning tricks scaling sgd</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:e4ea1d658f64/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:tricks"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sgd"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://adventuresindatascience.wordpress.com/2014/12/30/minibatch-learning-for-large-scale-data-using-scikit-learn/">
    <title>Minibatch learning for large-scale data, using scikit-learn | Adventures in Data Science</title>
    <dc:date>2015-02-16T01:46:34+00:00</dc:date>
    <link>https://adventuresindatascience.wordpress.com/2014/12/30/minibatch-learning-for-large-scale-data-using-scikit-learn/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>sklearn machine-learning scaling sgd online-learning mini-batch</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:2f65cd84829a/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sklearn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sgd"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:online-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mini-batch"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://jmlr.org/proceedings/papers/v32/gieseke14.html">
    <title>Buffer k-d Trees: Processing Massive Nearest Neighbor Queries on GPUs | ICML 2014 | JMLR W&amp;CP</title>
    <dc:date>2015-01-04T11:56:48+00:00</dc:date>
    <link>http://jmlr.org/proceedings/papers/v32/gieseke14.html</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["We present a new approach for combining k-d trees and graphics processing units for nearest neighbor search. It is well known that a direct combination of these tools leads to a non-satisfying performance due to conditional computations and suboptimal memory accesses. To alleviate these problems, we propose a variant of the classical k-d tree data structure, called buffer k-d tree, which can be used to reorganize the search. Our experiments show that we can take advantage of both the hierarchical subdivision induced by k-d trees and the huge computational resources provided by today’s many-core devices. We demonstrate the potential of our approach in astronomy, where hundreds of million nearest neighbor queries have to be processed."]]></description>
<dc:subject>papers kd-tree knn scaling gpu</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:3138bf9e45db/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:kd-tree"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:knn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:gpu"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1411.4000">
    <title>[1411.4000] How to Scale Up Kernel Methods to Be As Good As Deep Neural Nets</title>
    <dc:date>2014-12-16T21:21:37+00:00</dc:date>
    <link>http://arxiv.org/abs/1411.4000</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["In this paper, we investigate how to scale up kernel methods to take on large-scale problems, on which deep neural networks have been prevailing. To this end, we leverage existing techniques and develop new ones. These techniques include approximating kernel functions with features derived from random projections, parallel training of kernel models with 100 million parameters or more, and new schemes for combining kernel functions as a way of learning representations. We demonstrate how to muster those ideas skillfully to implement large-scale kernel machines for challenging problems in automatic speech recognition. We valid our approaches with extensive empirical studies on real-world speech datasets on the tasks of acoustic modeling. We show that our kernel models are equally competitive as well-engineered deep neural networks (DNNs). In particular, kernel models either attain similar performance to, or surpass their DNNs counterparts. Our work thus avails more tools to machine learning researchers in addressing large-scale learning problems."]]></description>
<dc:subject>papers speech asr dnn kernel-methods scaling acoustic-model</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:d021726895f1/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:speech"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:asr"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:dnn"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:kernel-methods"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:acoustic-model"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://static.googleusercontent.com/media/research.google.com/en/us/people/jeff/CIKM-keynote-Nov2014.pdf">
    <title>Large Scale Deep Learning</title>
    <dc:date>2014-12-10T17:19:50+00:00</dc:date>
    <link>http://static.googleusercontent.com/media/research.google.com/en/us/people/jeff/CIKM-keynote-Nov2014.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>slides jeff-dean scaling machine-learning deep-learning neural-net</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:9cb198d86966/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:slides"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:jeff-dean"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:deep-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://madalgo.au.dk/fileadmin/madalgo/Summer_School_2014/Introduction__2_.pdf">
    <title>Summer School 2014 Learning at Scale</title>
    <dc:date>2014-10-26T04:05:20+00:00</dc:date>
    <link>http://madalgo.au.dk/fileadmin/madalgo/Summer_School_2014/Introduction__2_.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>slides presentation machine-learning scaling .inspiration</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:a19acf5f5d46/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:slides"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:presentation"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:.inspiration"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://jsatml.blogspot.com/2013/02/mini-batch-k-means-vs-elkan.html">
    <title>Machine Learning with JSAT: Mini Batch K-Means vs Elkan</title>
    <dc:date>2014-10-09T04:25:21+00:00</dc:date>
    <link>http://jsatml.blogspot.com/2013/02/mini-batch-k-means-vs-elkan.html</link>
    <dc:creator>arsyed</dc:creator><dc:subject>algorithms clustering mini-batch scaling comparison k-means</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:6d6518d7012f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:clustering"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mini-batch"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:comparison"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:k-means"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://github.com/ptoulis/implicit-sgd">
    <title>ptoulis/implicit-sgd</title>
    <dc:date>2014-10-04T16:50:29+00:00</dc:date>
    <link>https://github.com/ptoulis/implicit-sgd</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["Using stochastic gradient descent (SGD) with explicit and implicit updates to fit large-scale statistical models."]]></description>
<dc:subject>R code sgd scaling big-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:9273163dcc74/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:R"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:code"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:sgd"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://curtis.ml.cmu.edu/w/courses/index.php/Machine_Learning_with_Large_Datasets_10-605_in_Spring_2014">
    <title>Machine Learning with Large Datasets 10-605 in Spring 2014 - Cohen Courses</title>
    <dc:date>2014-09-28T15:04:03+00:00</dc:date>
    <link>http://curtis.ml.cmu.edu/w/courses/index.php/Machine_Learning_with_Large_Datasets_10-605_in_Spring_2014</link>
    <dc:creator>arsyed</dc:creator><dc:subject>courses machine-learning performance mmds scaling big-data william-cohen cmu</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:5a91053fded8/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:courses"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:performance"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mmds"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:william-cohen"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:cmu"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://forty9ten.com/post/64676776664/avoiding-chef-suck-with-auto-scaling-groups">
    <title>Avoiding Chef-Suck with Auto Scaling Groups - forty9ten</title>
    <dc:date>2014-09-23T15:32:10+00:00</dc:date>
    <link>http://forty9ten.com/post/64676776664/avoiding-chef-suck-with-auto-scaling-groups</link>
    <dc:creator>arsyed</dc:creator><dc:subject>deployment chef ec2 scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:69c1b6667145/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:deployment"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:chef"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ec2"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://dl.acm.org/citation.cfm?id=1135870">
    <title>Large-scale text categorization by batch mode active learning</title>
    <dc:date>2014-09-03T15:03:00+00:00</dc:date>
    <link>http://dl.acm.org/citation.cfm?id=1135870</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Large-scale text categorization is an important research topic for Web data mining. One of the challenges in large-scale text categorization is how to reduce the human efforts in labeling text documents for building reliable classification models. In the past, there have been many studies on applying active learning methods to automatic text categorization, which try to select the most informative documents for labeling manually. Most of these studies focused on selecting a single unlabeled document in each iteration. As a result, the text categorization model has to be retrained after each labeled document is solicited. In this paper, we present a novel active learning algorithm that selects a batch of text documents for labeling manually in each iteration. The key of the batch mode active learning is how to reduce the redundancy among the selected examples such that each example provides unique information for model updating. To this end, we use the Fisher information matrix as the measurement of model uncertainty and choose the set of documents to effectively maximize the Fisher information of a classification model. Extensive experiments with three different datasets have shown that our algorithm is more effective than the state-of-the-art active learning techniques for text categorization and can be a promising tool toward large-scale text categorization for World Wide Web documents.
]]></description>
<dc:subject>papers active-learning batch-mode scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:282e8017d527/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:active-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:batch-mode"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://etda.libraries.psu.edu/paper/9484/">
    <title>Learning In Extreme Conditions: Online And Active Learning With Massive, Imbalanced And Noisy Data</title>
    <dc:date>2014-09-02T18:23:14+00:00</dc:date>
    <link>https://etda.libraries.psu.edu/paper/9484/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["This thesis addresses improving the performance of machine learning algorithms with a particular focus on classification tasks with large, imbalanced and noisy datasets. The field of machine learning addresses the question of how best to use experimental or historical data to discover general patterns and regularities and improve the process of decision making. However, applying machine learning algorithms to very large scale problems still poses challenges. Additionally, class imbalance and noise in the data degrade the prediction accuracy of standard machine learning algorithms. The main focus of this thesis is designing machine learning algorithms and approaches that are faster, data efficient and less demanding in computational resources to achieve scalable algorithms for large scale problems. This thesis addresses these problems in active and online learning frameworks. The particular focus of the thesis is on Support Vector Machine (SVM) algorithm with classification problems, but the proposed approaches on active and online learning are also well extensible to other widely used machine learning algorithms."]]></description>
<dc:subject>papers thesis machine-learning online-learning active-learning scaling svm seyda-ertekin unbalanced-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:eb0aba101dea/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:thesis"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:online-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:active-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:svm"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:seyda-ertekin"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:unbalanced-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5206651&amp;abstractAccess=no&amp;userType=inst">
    <title>Active learning for large multi-class problems</title>
    <dc:date>2014-09-02T18:14:36+00:00</dc:date>
    <link>http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5206651&amp;abstractAccess=no&amp;userType=inst</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Scarcity and infeasibility of human supervision for large scale multi-class classification problems necessitates active learning. Unfortunately, existing active learning methods for multi-class problems are inherently binary methods and do not scale up to a large number of classes. In this paper, we introduce a probabilistic variant of the K-nearest neighbor method for classification that can be seamlessly used for active learning in multi-class scenarios. Given some labeled training data, our method learns an accurate metric/kernel function over the input space that can be used for classification and similarity search. Unlike existing metric/kernel learning methods, our scheme is highly scalable for classification problems and provides a natural notion of uncertainty over class labels. Further, we use this measure of uncertainty to actively sample training examples that maximize discriminating capabilities of the model. Experiments on benchmark datasets show that the proposed method learns appropriate distance metrics that lead to state-of-the-art performance for object categorization problems. Furthermore, our active learning method effectively samples training examples, resulting in significant accuracy gains over random sampling for multi-class problems involving a large number of classes.

]]></description>
<dc:subject>papers active-learning scaling multi-class classification</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:0c0d12143c6f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:active-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:multi-class"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:classification"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.chioka.in/tea-time-with-convolutional-deep-belief-networks-for-scalable-unsupervised-learning-of-hierarchical-representations/">
    <title>Tea Time With: Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations | Garbled Notes</title>
    <dc:date>2014-08-03T06:42:14+00:00</dc:date>
    <link>http://www.chioka.in/tea-time-with-convolutional-deep-belief-networks-for-scalable-unsupervised-learning-of-hierarchical-representations/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["Reasons why deep learning algorithms do not scale:
[...]
RBM choke because the same feature detector learned at a location cannot be used on another location (i.e. not translation invariant).
[...]
A traditional Restricted Boltzmann Machine (RBM) has no pooling layer. The CRBM, which makes up the CDBN, has this extra pooling layer on top of hidden layer which shrinks the hidden layer details before feeding it to the next layer in the CDBN. Shrinking is to drop the details learnt by a factor or 2 or 3. This effectively drops some noises which allows small translations to happen, reducing computational burden as well.
In DBNs, when a given feature is learnt (i.e. the weights are learnt), it is not shared across other locations of the image. This redundancy means extra computational work. CDBNs have weights that are shared among all locations in an image."]]></description>
<dc:subject>neural-net convnet pooling scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:ceb4b7b6788d/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:convnet"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pooling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://bcb9f395-a-8ac90a7d-s-sites.googlegroups.com/a/vanhoucke.com/vincent/publications/vanhoucke-iclr14.pdf?attachauth=ANoY7coFEnIKISzOLVnX0s9zwv5-d15ocVtVDIqHbaIVBeqss0NoD9hk1FCvNXqNhjdi8LgmxxP-TqjEytBPFcWk91EG3f0WfvYYhfG1ftJq_8SmtTwVvwyn_aASMFY9ENoBoT5UlUv067EcBCi5_MHT2noyphZ9z_Oq0SJpLQOl8s8plVBpbgETF7ZkTf1GU2Mzll2AItk7GmOG9r8RhVYPTFN3mp35X4oML4yr3yFp4dVFA3-yKkw%3D&amp;attredirects=0">
    <title>Learning Visual Representations at Scale</title>
    <dc:date>2014-04-16T18:32:23+00:00</dc:date>
    <link>https://bcb9f395-a-8ac90a7d-s-sites.googlegroups.com/a/vanhoucke.com/vincent/publications/vanhoucke-iclr14.pdf?attachauth=ANoY7coFEnIKISzOLVnX0s9zwv5-d15ocVtVDIqHbaIVBeqss0NoD9hk1FCvNXqNhjdi8LgmxxP-TqjEytBPFcWk91EG3f0WfvYYhfG1ftJq_8SmtTwVvwyn_aASMFY9ENoBoT5UlUv067EcBCi5_MHT2noyphZ9z_Oq0SJpLQOl8s8plVBpbgETF7ZkTf1GU2Mzll2AItk7GmOG9r8RhVYPTFN3mp35X4oML4yr3yFp4dVFA3-yKkw%3D&amp;attredirects=0</link>
    <dc:creator>arsyed</dc:creator><dc:subject>slides neural-net deep-learning convnet representation-learning computer-vision scaling parallel</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:f0242062374f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:slides"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:neural-net"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:deep-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:convnet"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:representation-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:computer-vision"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.cs.ucsb.edu/~veronika/MAE_LargeScaleClustering_strnadova.pdf">
    <title>Clustering at Large Scales</title>
    <dc:date>2014-03-25T01:58:45+00:00</dc:date>
    <link>http://www.cs.ucsb.edu/~veronika/MAE_LargeScaleClustering_strnadova.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>clustering scaling algorithms surveys .* slides</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:d0df0ae2b59f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:clustering"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:surveys"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:.*"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:slides"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.mblondel.org/lightning/">
    <title>lightning — lightning dev documentation</title>
    <dc:date>2014-02-04T02:11:01+00:00</dc:date>
    <link>http://www.mblondel.org/lightning/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["lightning is a library for large-scale linear classification and regression in Python."

]]></description>
<dc:subject>python libs machine-learning classification regression scaling performance ranking</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:0ff9614dda03/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:python"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:libs"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:classification"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:regression"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:performance"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ranking"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.holovaty.com/writing/aws-notes/">
    <title>Why I left Heroku, and notes on my new AWS setup | Holovaty.com</title>
    <dc:date>2014-01-13T19:33:01+00:00</dc:date>
    <link>http://www.holovaty.com/writing/aws-notes/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA[Adrian Holovaty's post from May 2013.  Critique of Heroku which didn't work out well for Soundslice.  Description of his setup on AWS with baked in scaling, redundancy which he's happy about.  He switched from Postgresql to MySQL for RDS.]]></description>
<dc:subject>heroku django deployment amazon aws scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:597822851e20/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:heroku"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:django"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:deployment"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:amazon"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:aws"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://elelsee.com/">
    <title>elelsee</title>
    <dc:date>2014-01-13T19:30:17+00:00</dc:date>
    <link>http://elelsee.com/</link>
    <dc:creator>arsyed</dc:creator><dc:subject>people webdev scaling devops ofa obama</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:94ec4ff6ea2d/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:people"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:webdev"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:devops"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:ofa"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:obama"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://www.jstatsoft.org/v55/i14">
    <title>Scalable Strategies for Computing with Massive Data</title>
    <dc:date>2014-01-11T00:35:17+00:00</dc:date>
    <link>http://www.jstatsoft.org/v55/i14</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["This paper presents two complementary statistical computing frameworks that address challenges in parallel processing and the analysis of massive data. First, the foreach package allows users of the R programming environment to define parallel loops that may be run sequentially on a single machine, in parallel on a symmetric multiprocessing (SMP) machine, or in cluster environments without platform-specific code. Second, the bigmemory package implements memory- and file-mapped data structures that provide (a) access to arbitrarily large data while retaining a look and feel that is familiar to R users and (b) data structures that are shared across processor cores in order to support efficient parallel computing techniques. Although these packages may be used independently, this paper shows how they can be used in combination to address challenges that have effectively been beyond the reach of researchers who lack specialized software development skills or expensive hardware."]]></description>
<dc:subject>R parallel memory scaling</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:466e1f863222/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:R"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:parallel"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:memory"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="https://code.facebook.com/posts/218678814984400/scaling-mercurial-at-facebook/">
    <title>Scaling Mercurial at Facebook | Engineering Blog | Facebook Code | Facebook</title>
    <dc:date>2014-01-08T06:04:24+00:00</dc:date>
    <link>https://code.facebook.com/posts/218678814984400/scaling-mercurial-at-facebook/</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["A big benefit of distributed source control is the ability to work without interacting with the server. The remotefilelog extension intelligently caches the file revisions needed for your local commits so you can checkout, rebase, and commit to any of your existing bookmarks without needing to access the server. Since we still download all of the commit metadata, operations that don't require file contents (such as log) are completely local as well. Lastly, we use Facebook's memcache infrastructure as a caching layer in front of the central Mercurial server, so that even if the central repository goes down, memcache will continue to serve many of the file content requests."

]]></description>
<dc:subject>facebook mercurial dvcs scaling via:chl performance</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:81a3caa64acd/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:facebook"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mercurial"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:dvcs"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:via:chl"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:performance"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1110.4198">
    <title>[1110.4198] A Reliable Effective Terascale Linear Learning System</title>
    <dc:date>2013-11-10T20:07:08+00:00</dc:date>
    <link>http://arxiv.org/abs/1110.4198</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["We present a system and a set of techniques for learning linear predictors with convex losses on terascale datasets, with trillions of features, {The number of features here refers to the number of non-zero entries in the data matrix.} billions of training examples and millions of parameters in an hour using a cluster of 1000 machines. Individually none of the component techniques are new, but the careful synthesis required to obtain an efficient implementation is. The result is, up to our knowledge, the most scalable and efficient linear learning system reported in the literature (as of 2011 when our experiments were conducted). We describe and thoroughly evaluate the components of the system, showing the importance of the various design choices."]]></description>
<dc:subject>papers machine-learning algorithms scaling big-data</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:a04e03eb8268/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:algorithms"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:big-data"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://arxiv.org/abs/1112.5016">
    <title>[1112.5016] A Scalable Bootstrap for Massive Data</title>
    <dc:date>2013-09-08T02:09:06+00:00</dc:date>
    <link>http://arxiv.org/abs/1112.5016</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["The bootstrap provides a simple and powerful means of assessing the quality of estimators. However, in settings involving large datasets---which are increasingly prevalent---the computation of bootstrap-based quantities can be prohibitively demanding computationally. While variants such as subsampling and the m out of n bootstrap can be used in principle to reduce the cost of bootstrap computations, we find that these methods are generally not robust to specification of hyperparameters (such as the number of subsampled data points), and they often require use of more prior information (such as rates of convergence of estimators) than the bootstrap. As an alternative, we introduce the Bag of Little Bootstraps (BLB), a new procedure which incorporates features of both the bootstrap and subsampling to yield a robust, computationally efficient means of assessing the quality of estimators. BLB is well suited to modern parallel and distributed computing architectures and furthermore retains the generic applicability and statistical efficiency of the bootstrap. We demonstrate BLB's favorable statistical performance via a theoretical analysis elucidating the procedure's properties, as well as a simulation study comparing BLB to the bootstrap, the m out of n bootstrap, and subsampling. In addition, we present results from a large-scale distributed implementation of BLB demonstrating its computational superiority on massive data, a method for adaptively selecting BLB's hyperparameters, an empirical study applying BLB to several real datasets, and an extension of BLB to time series data."]]></description>
<dc:subject>bagging via:arthegall papers statcomp bootstrap scaling</dc:subject>
<dc:identifier>https://pinboard.in/u:arsyed/b:ca8992355346/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:bagging"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:via:arthegall"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:papers"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:statcomp"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:bootstrap"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://stats.stackexchange.com/questions/63589/how-to-project-high-dimensional-space-into-a-two-dimensional-plane">
    <title>data visualization - How to project high dimensional space into a two-dimensional plane? - Cross Validated</title>
    <dc:date>2013-07-14T03:19:58+00:00</dc:date>
    <link>http://stats.stackexchange.com/questions/63589/how-to-project-high-dimensional-space-into-a-two-dimensional-plane</link>
    <dc:creator>arsyed</dc:creator><description><![CDATA["The key point is not that MDS is for distances input and PCA is for coordinates input, but that iterative MDS fits few dimensions while PCA retains few dimensions. So MDS preserves distances somewhat better than classic PCA does. The answer for the question is Yes, PCA is suitable, but MDS is more suitable."]]></description>
<dc:subject>embedding mds pca scaling dimensionality-reduction</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:cf410e4ce54c/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:embedding"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:mds"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:pca"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:dimensionality-reduction"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://cseweb.ucsd.edu/~akmenon/ResearchExam.pdf">
    <title>Large-Scale Support Vector Machines: Algorithms and Theory [pdf]</title>
    <dc:date>2013-04-26T18:35:37+00:00</dc:date>
    <link>http://cseweb.ucsd.edu/~akmenon/ResearchExam.pdf</link>
    <dc:creator>arsyed</dc:creator><dc:subject>machine-learning svm scaling bigdata</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:f7695005c49a/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:machine-learning"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:svm"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:bigdata"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://news.ycombinator.com/item?id=5215884">
    <title>Heroku's Ugly Secret: The story of how the cloud-king turned its back on Rails | Hacker News</title>
    <dc:date>2013-02-14T12:10:50+00:00</dc:date>
    <link>http://news.ycombinator.com/item?id=5215884</link>
    <dc:creator>arsyed</dc:creator><dc:subject>queueing load-balancing scaling heroku</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:e55590b9cc7a/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:queueing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:load-balancing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:heroku"/>
</rdf:Bag></taxo:topics>
</item>
<item rdf:about="http://rapgenius.com/James-somers-herokus-ugly-secret-lyrics">
    <title>James Somers – Heroku's Ugly Secret | Rap Genius</title>
    <dc:date>2013-02-14T12:10:23+00:00</dc:date>
    <link>http://rapgenius.com/James-somers-herokus-ugly-secret-lyrics</link>
    <dc:creator>arsyed</dc:creator><dc:subject>heroku scaling load-balancing queueing</dc:subject>
<dc:source>https://pinboard.in/</dc:source>
<dc:identifier>https://pinboard.in/u:arsyed/b:33cc45b7a05f/</dc:identifier>
<taxo:topics><rdf:Bag>	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:heroku"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:scaling"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:load-balancing"/>
	<rdf:li rdf:resource="https://pinboard.in/u:arsyed/t:queueing"/>
</rdf:Bag></taxo:topics>
</item>
</rdf:RDF>