Robert Bamler(University of California, Irvine)
hosted by Department of Computer Science
"Resource-Efficient Bayesian Machine Learning"
Recent success stories of Machine Learning in applications likesearch, voice assist, and autonomous driving are sparking curiosity from manyother researchers and engineers to apply Machine Learning methods in theirrespective fields. However, a lot of the recent progress in Machine Learning hasbeen fueled by heavy investments into enormous data sets and computing power.Such resource hungry methods are not viable in many areas, such as naturalsciences, Internet of Things, or emerging decentralized computing platforms. Inthis talk, I will propose a path forward for the development of more resourceefficient Machine Learning methods. I will address efficiency in training data,communication bandwidth, storage, and computing power. Using examples from ownresearch, I will make the case that recent advances in scalable methods forapproximate Bayesian inference provide a unifying framework to address all theseaspects of efficiency, especially if combined with knowledge from other areas ofcomputer or natural science.
|Time:||Monday, 02.03.2020, 14:00|