Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session R51: Artificial Intelligence, Data, and Dynamics: Learning Physical Models of Living SystemsInvited Undergraduate
|
Hide Abstracts |
Sponsoring Units: DBIO Chair: Ilya Nemenman, Emory University Room: BCEC 253A |
Thursday, March 7, 2019 8:00AM - 8:36AM |
R51.00001: Machine Learning, Statistical Physics, and Ecological Dynamics Invited Speaker: Pankaj Mehta In this talk, I will start by giving an overview of Machine Learning from a physics perspective and highlight open problems where physicists can contribute. I will then discuss the many connections between the statistical physics of disordered systems and ML. Building on this discussion, I will argue that, somewhat suprisingly, ML is also intimately related to ecological dynamics. I will show how many ML methods and concepts have natural counterparts and ecology and argue that these fields can and should cross-fertalize each other. |
Thursday, March 7, 2019 8:36AM - 9:12AM |
R51.00002: Principles and Possibilities in the Phase Space of Animal Behavior Invited Speaker: Greg Stephens We all instinctively recognize behavior: it’s what organisms do, from the motility of single cells to the stunning displays of bird flocks. To understand behavior, however, we must characterize complex, living movement as precisely and completely as its underlying molecular, cellular and network mechanisms. Here, we leverage a low-dimensional but complete representation of the posture of nematode worm C. elegans to reconstruct a continuous 6D phase space of crawling behavior. The reconstruction separates short and long-time dynamics, untangles subtle movement trajectories, and offers a quantitative arena for examining variability and stereotypy. We find that the phase space is organized into 3 conjugate dynamics containg 2 positive Lyapunov exponents which are approximately balanced by dissipative directions. We suggest that a near-Hamiltonian dynamics of coupled, chaotic oscillators underlie the motor control of C. elegans. |
Thursday, March 7, 2019 9:12AM - 9:48AM |
R51.00003: The difference between memory and prediction in recurrent networks Invited Speaker: Sarah Marzen Recurrent networks are trained to memorize their input better, often in the hopes that such training will increase the ability of the network to predict. We show that networks designed to memorize input can be arbitrarily bad at prediction. We also find, for several types of inputs, that one-node networks optimized for prediction are nearly at upper bounds on predictive capacity given by Wiener filters, and are roughly equivalent in performance to randomly generated five-node networks. Our results suggest that maximizing memory capacity leads to very different networks than maximizing predictive capacity. We also discuss how well trained recurrent networks can predict, compared to the optimal. |
Thursday, March 7, 2019 9:48AM - 10:24AM |
R51.00004: Hybrid Forecasting of Complex Systems: Combing Machine Learning with Knowledge-based Models Invited Speaker: Michelle Girvan In recent years, machine learning methods such as "deep learning" have proven enormously successful for tasks such as image classification, voice recognition, and more. Despite their effectiveness for big-data classification problems, these methods have had limited success for time series prediction, especially for complex systems like those we see in weather, solar activity, and brain dynamics. In this talk, I will discuss how a Reservoir Computer (RC) - a special kind of machine learning system that offers a "universal" dynamical system - can draw on its own internal complex dynamics in order to forecast systems like the weather, beyond the time horizon of other methods. The RC provides a knowledge-free approach because it builds forecasts purely from past measurements without any specific knowledge of the system dynamics. By building a new hybrid approach that judiciously combines the knowledge-free prediction of the RC with a knowledge-based, mechanistic model, we demonstrate a further, dramatic, improvement in forecasting complex systems. This hybrid approach can given us new insights into the weaknesses of our knowledge-based models and also reveal limitations in our machine learning system, guiding improvements in both knowledge-free and knowledge-based prediction techniques. |
Thursday, March 7, 2019 10:24AM - 11:00AM |
R51.00005: Measuring the hidden dynamics of animal behavior Invited Speaker: Gordon Berman When we think of animal behavior, what typically comes to mind are actions – running, eating, swimming, grooming, flying, singing, resting. Behavior, however, is more than the catalogue of motions that an organism can perform. Animals organize their repertoire of actions into sequences and patterns whose underlying dynamics last much longer than any particular behavior. How an organism modulates these dynamics affects its success at accessing food, reproducing, and myriad other tasks essential for survival. Animals regulate these patterns of behavior via many interacting internal states (hunger, reproductive cycle, age, etc.) that we cannot directly measure. Studying these hidden states’ dynamics, accordingly, has proven challenging due to a lack of measurement techniques and theoretical understanding. In this talk, I will outline our efforts to uncover the latent dynamics that underlie long timescale structure in animal behavior. Looking across a variety of organisms, we find the existence of a non-trivial form of long timescale dynamics that is unexplainable in standard Markovian frameworks. I will present how temporal coarse-graining can be used to understand how these dynamics are generated and how the found course-grained states can be related to internal states governing behavior. Inferring these hidden dynamics presents a new opportunity to generate insights into the neural and physiological mechanisms that animals use to select actions. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700