How We Share Machine Learning Knowledge at Microsoft

  • Article

We recently concluded the Fall 2014 edition of our Practice of Machine Learning Conference (PMLC). Over 1,700 Microsoft employees attended the two day event, which featured 60 talks on a broad spectrum of areas ranging from new algorithms to ML applications such as anomaly detection and fraud. Tutorials covered such topics as feature engineering, labeling, Azure Machine Learning, multi-world testing and more. A Poster & Demo Reception saw more than 50 presenters. And our first-ever Azure ML Cloud App Contest featured many interesting contenders.

To share a flavor of the event with you, we boiled our 2 day event into a 2 minute video:

 

The best part was that ML practitioners from dozens of teams and locations around the world got a chance to connect in person, discuss the latest advances in ML and learn from one another. Presenters from Beijing, Cambridge, Herzelia, Hyderabad and 15 other cities flew in to participate in person and share their work. Some of the most popular sessions included:

  • Automatic Image Captioning at a Human Level of Performance

  • Beating Paul the Octopus: How Bing Uses Web and Social Data for Predictive Modeling

  • Customer Churn Prediction Service on Azure ML

  • Exploring ML with scikit-learn

  • Introduction to Fireworks Algorithm Optimization

  • Real-Time Stream Analytics Using Azure Stream Analytics

At his keynote talk, Joseph Sirosh, Corporate Vice President of Information Management & Machine Learning (IMML), talked about the newly emerging Data Science Economy and the opportunities afforded by Azure ML for data scientists, including the ability to self-publish and monetize their skills through the Azure Marketplace in a manner similar to how developers monetize their skills through app stores today. This echoed his message at the recent Strata + Hadoop 2014 event which had a similar theme.

In the first-ever Azure ML Cloud App Contest, participants built Azure ML apps of their choice that addressed real-world scenarios. The winning app, Dr. Pig, created by a Microsoft team in China, helps small-scale pig breeders forecast future prices and profits and mitigate risk by helping with decisions such as the type of pig to breed.

The conference concluded with a closing panel designed to stimulate spirited discussion – in “Are We at Peak ML? Hype vs. Reality of Machine Learning”, John Platt, Distinguished Scientist at Microsoft Research, and Greg Buehrer, Partner Development Manager, joined Joseph to debate the future directions of ML.

The PMLC is held twice a year. Between conferences, the Microsoft ML community gathers regularly for smaller in-person and online events where practitioners share their experience and expertise. Members also participate in ML-focused hackathons and internal forums, and have access to resources such as the online Machine Learning University, which pulls together course materials on the most-requested ML topics. These activities are designed to allow ML practitioners from around the world share their knowledge, help one another and make Microsoft products and services better through the power of ML.

To learn more about ML at Microsoft, you can subscribe to our ML blog feed or follow us on Twitter @MLatMSFT. And, for those of you practitioners who did not know yet, we recently made Azure ML available free of charge without a subscription or credit card – just click the blue “Get Started Now” button at the Azure ML Preview page and get going today.

ML Blog Team