MLconf Industry Impact Student Research Award Winners

This year, we started a new award program called the MLconf Industry Impact Student Research Award, which is sponsored by Google. This fall, our committee of distinguished ML professionals reviewed several nominations sent in from members of the MLconf community. There were several great researchers that were nominated and the committee arrived at awarding 2 students whose work, they believe, has the potential to disrupt the industry in the future. The two winners that were announced at MLconf SF 2015 are UC Irvine Student, Furong Huang and UC Berkeley Student, Virginia Smith. Below are summaries of their research. We’ve invited both researchers to present their work at upcoming MLconf events.

Furong Huang, Research Assistant, UC Irvine

Furong Huang’s research has focused on tensor factorization models and their application in unsupervised learning. There are three key observations that make her research successful. The first observation is the information richness of higher order moment tensors. By using matrices it was only possible to look at second order moments. With tensors it is possible to look at higher order moments. The second observation has to do with the fact that the elements of those decomposed tensors estimate parameters for many hard machine learning problems, such as mixture of experts, latent models, mutli layer networks etc. So the tensors can be a unified learning platform. The third observation, is that the tensor decomposition although a non-convex optimization problem, all the local minima can be equivalent. This is a key advantage of the whole framework as in the industry it is very important to have consistent results. It makes debugging easy. Another reason why we believe the tensor learning framework has the potential to grow, is its simplicity and scalability.The framework convert the complex high-dimensional learning problem to a simple tensor decomposition problem in a low-dimensional and compact space. Therefore, the framework is scalable to data in high-dimensions. At last we need to point out that the mapping of the problem to high-level BLAS linear algebra operations makes tensor learning extremely fast with the use of GPUs and other optimized linear algebra HPC packages. Furong has taken the extra mile to implement her work in different platforms (GPU, Apache Spark and Hadoop MapReduce) that are widely accepted by the industry.

Virginia Smith, Researcher, UC Berkeley

Virginia Smith’s research focuses on distributed optimization for large-scale machine learning. The main challenge in many large-scale machine learning tasks is to solve an optimization objective involving data that is distributed across multiple machines. In this setting, optimization methods that work well on single machines must be re-designed to leverage parallel computation while reducing communication costs. This requires developing new distributed optimization methods with both competitive practical performance and strong theoretical convergence guarantees. Virginia’s work aims to determine policies for distributed computation that meet these requirements, in particular through the development of a novel primal-dual framework, CoCoA, which is written on Spark. The theoretical and practical development of CoCoA is an important step for future data scientists hoping to deploy efficient large-scale machine learning algorithms.