<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>ml on Max Hoffman</title>
    <link>https://hoffman.ai/tags/ml/</link>
    <description>Recent content in ml on Max Hoffman</description>
    <generator>Hugo -- gohugo.io</generator>
    <language>en</language>
    <lastBuildDate>Fri, 05 Jun 2020 10:51:04 -0700</lastBuildDate>
    
	<atom:link href="https://hoffman.ai/tags/ml/index.xml" rel="self" type="application/rss+xml" />
    
    
    <item>
      <title>Lambda MR</title>
      <link>https://hoffman.ai/posts/lambda-mr/</link>
      <pubDate>Fri, 05 Jun 2020 10:51:04 -0700</pubDate>
      
      <guid>https://hoffman.ai/posts/lambda-mr/</guid>
      <description>What problems to Spark and Map Reduce solve? Was Spark the next generation of MapReduce? Do they solve the same problem? Is Spark quick and failure-prone, while MapReduce is slow and reliable?
To play devil&amp;rsquo;s advocate, I think MapReduce was the right abstraction for the wrong problem.</description>
    </item>
    
    <item>
      <title>Smaker</title>
      <link>https://hoffman.ai/posts/smaker/</link>
      <pubDate>Sat, 25 May 2019 15:28:52 -0700</pubDate>
      
      <guid>https://hoffman.ai/posts/smaker/</guid>
      <description>Smaker Workflow Helper Motivation My second project at Factual involved re-evaluating our team&amp;rsquo;s workflow automation software. The existing system used Drake to run our pre-processing, training and evalutation scripts. Drake introduces many pain-points in the process of writing and maintaining workflows, but I was more concerned with the lack of certain features that I considered important for our use cases:</description>
    </item>
    
    <item>
      <title>Graph Think with WAVE</title>
      <link>https://hoffman.ai/posts/wave/wave/</link>
      <pubDate>Tue, 08 Jan 2019 14:52:05 -0600</pubDate>
      
      <guid>https://hoffman.ai/posts/wave/wave/</guid>
      <description>I spent some time working with graph recurrent neural networks at WashU. One of the key motivations of the project was to design a system that played to a graph&amp;rsquo;s strengths, rather reusing models designed for different applications.
As a lab engineer, I did not own any of the science, so I did not spend the bulk of my time trying to understand the literature or design careful experiements.</description>
    </item>
    
    <item>
      <title>Tflon</title>
      <link>https://hoffman.ai/posts/tflon/</link>
      <pubDate>Sun, 30 Sep 2018 06:55:06 -0500</pubDate>
      
      <guid>https://hoffman.ai/posts/tflon/</guid>
      <description>The Swamidass lab I am researching with has a tooling framework called Tflon. Matthew Mattlock, who wrote most of Tflon, first described the project as his attempt to provide tooling that shared Keras&amp;rsquo;s ease of use, but allowed for the full flexibiliy of Tensorflow.</description>
    </item>
    
  </channel>
</rss>