Why Systems Biology Shouldn’t Work but Does and What Heat Capacity Can Explain About Inference
Speaker: Paul Wiggins, University of Washington
Why do systems biology models work in spite of a blizzard of poorly-defined parameters and yet the detection of the Higgs boson required five sigma precision? Scientific and technological innovations are rapidly increasing the size and scope of datasets. Accompanying this growth come new challenges in analysis, interpretation and modeling. Fundamental questions remain about the mechanism of learning. To study the universal principles governing these processes, we expand upon a long-discussed correspondence between thermodynamics and statistics. This correspondence to thermal physics provides some surprising insights into the mechanism of learning. An analogy to heat capacity demonstrates both a universal scaling of learning algorithms as well as how and why these scaling relations fail in many of the most interesting models, including systems biology models. An analogy to the Gibbs entropy provides a new algorithm for efficient inference, well-suited to single-molecule-fluorescence measurements where the number of photons collected is small.
Host: Tristan Ursell
Date: Thursday, April 19th, 2018
Location: 100 Willamette Hall
Catered Reception: 3:40pm-3:55pm, Willamette Hall, Paul Olum Atrium