Why Systems Biology Shouldn’t Work… but Does… and What Heat Capacity Can Explain about Learning
University of Washington
Despite an intensifying interest in applications of machine learning to the analysis of big data, fundamental questions remain about the mechanisms of learning: (i) How can immensely complex models ever learn from small datasets? (ii) What is the physics of learning and (iii) are there universal properties in learning processes? In this talk, I will elaborate on a long-discussed analogy between Bayesian statistics and statistical mechanics. This correspondence reveals a surprisingly simple answer to these three questions by analogy, in the well known physics of the heat capacity. Finally, I will discuss how these insights can be used to design new learning algorithms.
Monday, April 8, 2019 at 3:30pm to 4:30pm
Science & Engineering Building, 307
1414 Circle Dr, Knoxville, TN 37996