Machine Learning for Systems

Date(s):

Location:
Jacobs Hall, Room 2315, Jacobs School of Engineering, 9500 Gilman Dr, La Jolla, San Diego, California 92093

Sponsored By:
Professor Koushanfar

Speaker(s):
Dr. Azalia Mirhoseini
Google Brain
Dr. Azalia Mirhoseini

Abstract:

The recent success of machine learning has been driven by advances in computer systems, and now it is time for a new era in which computer systems design is transformed through machine learning. This talk will focus on two of our recent works: Resource Allocation Optimization with Deep Reinforcement Learning (RL) and Dynamic Neural Networks with Sparsely Gated Mixture of Experts.

The first half of the talk covers our new RL-based techniques to solve combinatorial optimization problems in the context of graph resource allocation. We show that our approach performs model parallelism without explicit profiling of the target hardware or the computational graph. Instead, it solves the problem by considering only the reward function of interest (e.g., runtime) and finds solutions that outperform traditional white box baselines.

 The second half of the talk covers our work on dynamic networks with sparse gates, where input examples are conditionally routed through the model. Dynamic networks allow us to performantly train models with much larger capacities. Combined with their implicit regularization properties, these models can yield significantly higher learning accuracies. Meanwhile, the sparse usage of the model architecture improves systems performance at both training and inference time, as shown by our state-of-the-art results on language modeling and machine translation benchmarks.


Speaker Bio:
Azalia Mirhoseini is a Research Scientist at Google Brain where she focuses on machine learning solutions to problems in computer systems and metalearning. Her research contributions include scalable reinforcement learning techniques to solve combinatorial optimization problems as well as a new data, algorithm, and systems co-design paradigm for high-performance machine learning. Before Google, she was at Rice University, where she got a Ph.D. in Electrical and Computer Engineering. Her work has been published at several top-tier conferences, including ICML, ICLR, DAC, ICCAD, and SIGMETRICS. She has received a number of awards, including the Best Ph.D. Thesis Award at Rice, fellowships from Microsoft Research, IBM Research, Schlumberger, and a Gold Medal in the National Math Olympiad in Iran.

Contact:
Wyn Hughes
whughes@eng.ucsd.edu
858-534-3294