BT

Facilitating the Spread of Knowledge and Innovation in Professional Software Development

Write for InfoQ

Topics

Choose your language

InfoQ Homepage Podcasts John Langford on Vowpal Wabbit, Used by MSN, and Machine Learning in Industry

John Langford on Vowpal Wabbit, Used by MSN, and Machine Learning in Industry

Bookmarks

In this week's podcast, QCon chair Wesley Reisz talks to Machine learning research scientist John Langford.  Topics discussed include his Machine Learning system Vowpal Wabbit, designed to be very efficient and incorporate some of the latest algorithms in the space.  Vowpal Wabbit is used for news personalisation on MSN.  They also discuss how to get started in the field and its shift from academic research to industry use.

Key Takeaways

  • Vowpal Wabbit is a ML system that attempts to incorporate some of the latest machine learning algorithms.
  • How to learn ML: take a class or two, get accustomed with learning theory and practice.
  • ML has moved from the research field into the industry, 4 out of 9 ICML tutorials coming from the industry.
  • It’s hard to predict when you have enough data.
  • AlphaGo is a milestone in artificial intelligence. It uses reinforcement learning, deep learning and existing moves played by Go masters.
  • Deep Learning is currently a disruptive technology in areas such a vision or speech recognition.
  • What’s trendy: Neural Networks, Reinforcement and Contextual Learning.
  • Machine Learning is being commoditized.

Notes

A Few Words about Vowpal Wabbit

  • 1m:38s - Vowpal Wabbit (VW) aims to incorporate the latest machine learning researches into algorithms.
  • 2m:04s - One algorithm is predicting one of k things, which can become computationally intensive if k is a large number. This Recall Tree algorithm is to be included in VW.
  • 3m:08s - VW can swallow 100MB of data in less than a sec. on a single machine, and the default learning algorithm is linear, being more powerful than Naive Bayse.
  • 3m:32s - VW uses an “assembly line of examples” to speed up parsing and improve processing time.
  • 4m:04s - At QCon New York, Langford’s talk introduced VW along with some examples demonstrating how it works.
  • 4m:25s - VW is used among others for a decision service for a personalized news recommendation system.
  • 4m:34s - The news recommendation system built on VW is deployed on MSN, which led to a 25% improvement in reader engagement.
  • 4m:58s - Vowpal Wabbit is actually how Elmer Fudd pronounces Vorpal Rabbit. Vorpal means very sharp and it comes from Jabberwocky, a poem written with nonsense words and understood because of the ways they are used. Then is the Killer Rabbit movie, where some people search for the Holy Grail and meet a particular rabbit.

How to Approach Machine Learning

Machine Learning in Practice

  • 7m:38s - Machine Learning is used a lot more in the industry than it was a decade ago.
  • 7m:58s - At the ICML machine learning conference, participation from industry is high with 4/9 tutorials coming from industry. And those are the most popular tutorials.
  • 8m:35s - Machine Learning is key for certain industries such as search or ads.
  • 9m:00s - Machine Learning is used for control, such as deciding when to plant or water crops with very good results.

When Enough Data is Enough

  • 12m:23s - It’s important to choose a representation and the features to be fed into a learning algorithm.
  • 13m:00s - It is hard to say when you have enough data. You try to model on the data you have, then add more data and see how much improvement there was, but at some point you run out of time or budget, and you settle for what you have.

AI and Go

  • 14m:23s - AlphaGo winning against the best human player in one of the most difficult games is a machine learning and AI milestone. It is a marker of progress in AI rather than a solution for AI.
  • 15m:02s - AlphaGo uses a combination of known techniques to achieve the result. It uses the Monte Carlo tree search which randomly picks a possible move out of the many existing ones. It is a fast algorithm.
  • 16m:54s - AlphaGo uses reinforcement learning, deep learning and existing moves played by Go masters.

Deep Learning

  • 17m:39s - Deep Learning is currently a disruptive technology in areas such a vision or speech recognition.
  • 17m:50s - Deep Learning won an image recognition competition in 2012, and it has been improving year after year. It has come to the error level of a human, able to recognize certain images about the same as humans can.

Trends

  • 9m:36s - 56 out of 332 papers at ICML were on Neural Networks, showing the importance of the domain. Reinforcement Learning and Optimization are also strong.
  • 18m:37s - There will be new deep learning applications.
  • 18m:50s - Dealing with causality, reinforcement and contextual learning will be important.  
  • 20m:20s - We are close to having machine learning as a commoditization service. Amazon, Google, Microsoft already provide machine learning services.

People/Companies Mentioned

Products Mentioned

More about our podcasts

You can keep up-to-date with the podcasts via our RSS Feed, and they are available via SoundCloud, Apple Podcasts, Spotify, Overcast and the Google Podcast. From this page you also have access to our recorded show notes. They all have clickable links that will take you directly to that part of the audio.

Previous podcasts

Rate this Article

Adoption
Style

Hello stranger!

You need to Register an InfoQ account or or login to post comments. But there's so much more behind being registered.

Get the most out of the InfoQ experience.

Allowed html: a,b,br,blockquote,i,li,pre,u,ul,p

Community comments

Allowed html: a,b,br,blockquote,i,li,pre,u,ul,p

Allowed html: a,b,br,blockquote,i,li,pre,u,ul,p

BT