Director, Nonlinear Science Program
Department of Physics
University of California, San Diego
Data Assimilation and Machine Learning as Statistical Physics Problems: Uses in Neuroscience and Deepest Learning
Many problems in Neurobiology and the Biology of Neural Systems require transferring sparse information from noisy laboratory observations to models of the complex, nonlinear systems producing the observations. We will discuss this from a Statistical Physics viewpoint using several examples, including: (a) completing and validating models of individual neurons using in vitro experiments, (b) determining the characteristics of neuromorphic VLSI chips representing neural behavior, and (c) designing experiments using calibrated neurons as network sensors in vivo to understand functional biological networks. The last item will use the equivalence of the Statistical Physics methods to supervised machine learning. An example of sequence learning from birdsong and an example of image classification from insect olfaction will be discussed.
A pizza lunch will be served.