cis6930fa11_ML_on_Multicore

cis6930fa11_ML_on_Multicore - Map-Reduce for Machine...

Info iconThis preview shows pages 1–7. Sign up to view the full content.

View Full Document Right Arrow Icon
Map-Reduce for Machine Learning on Multicore Chu, et al. Tuesday, August 30, 11
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Problem The world is going multicore New computers - dual core to 12+-core Shift to more concurrent programming paradigms and languages Erlang, Scala, Occam, Go, among others Tuesday, August 30, 11 - Paper discusses other languages, but 1) I dont know any of them 2) These languages are more up to date and useable.
Background image of page 2
Machine Learning frameworks on multicore How to parallelize algorithms? Highly specialized, non-obvious solutions Often applicable to very few algorithms Tuesday, August 30, 11 Example 1: Cargea et al - Restricted to decision trees Example 2: Jin, Agrawal - Shared Memory Machines only
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Develop general techniques for parallelizing many machine learning algorithms Goal of “throw more cores at it” optimization Algorithms must Ft the Statistical Query Model Goal Tuesday, August 30, 11 - Rather than traditional approach of algorithm speciFc tweaks
Background image of page 4
Statistical Query Model (Kearns, 1999) Permit learning algorithm to access learning problem only through a Statistical Oracle Oracle returns estimated expectation of f(x,y), given over all data instances Tuesday, August 30, 11 - The expectation averaged over the training/test distribution
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
All algorithms that computes sufFcient statistics can be expressed as a summation over all data points This trait lends itself to batch summations over all data points Why do we care? Tuesday, August 30, 11 Su f cient statistics - A statistic is sufFcient for a family of probability distributions if the sample from which it is calculated gives no additional information than does the statistic, as to which of those probability distributions is that of the population from which the sample was taken. Pr(X=x | T(x) = t, Θ ) = Pr(X=x | T(X) = t) conditional probability not reliant on underlying parameter Θ . (Source wikipedia article)
Background image of page 6
Image of page 7
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 27

cis6930fa11_ML_on_Multicore - Map-Reduce for Machine...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online