Congratulations to ECSE’s Tianyi Chen for receiving an Amazon Research Award for his proposal “Automating Decentralized Machine Learning via Bilevel Optimization.” The Amazon Research Awards is a program that provides unrestricted funds and AWS Promotional Credits to academic researchers investigating research topics across a number of disciplines.
Below is the abstract for Prof. Chen’s proposal:
Decentralized machine learning (DML) is a powerful paradigm that allows parallel and distributed computation within a data center and enables privacy-preserving distributed machine learning beyond data centers. However, conventional designs of DML algorithms are cumbersome since they involve many more hyper-parameters than their centralized counterpart such as stepsize, update frequency, participating clients, network topology and decentralized optimizer. This project aims to leverage bilevel optimization toolboxes to advance the automatic design of DML algorithms and its performance analysis. The new framework that we term Bilevel OptimizatiOn for decentralized Machine learning (BOOM) will generate both the conventional model-based decentralized optimization algorithms as well as the emerging model-free optimization algorithms that are unrolled in a deep neural network. BOOM will scale up centralized AI automation approaches in Amazon such as AutoGluon and Amazon SageMaker. Upon completion, the expected outcome includes open-source toolboxes that can potentially be integrated into the SageMaker platform and will expand the AutoML functions of SageMaker to decentralized settings.
Rama Hamarneh, email@example.com