Gerald Pao | Algorithms to map neural activity to Behavior
Foresight Institute’s Neurotech Group
Algorithms to map neural activity to Behavior
Bio: Dr. Gerald Pao worked originally on the molecular evolution of proteins from a structural and computational perspective as an undergraduate at the University of California, San Diego working with Milton H. Saier, Joseph Kraut, Flossie Wong-Staal, Russell F Doolittle and Tony Hunter. From there he went on to be mainly an experimentalist to study the epigenetics of cancer and stem cells and the development of viral vectors for basic science at the Salk institute during his PhD and postdoc with Inder M Verma. Work on stem cells led him to study the axolotl (Ambystoma mexicanum) salamander during limb regeneration at the Salk Institute and UC Irvine with David Gardiner and Tony Hunter. This was followed by a change of field through a postdoctoral training period in applied mathematics and data science specializing in nonlinear dynamics at the Scripps Institution of Oceanography (SIO) in the climate atmospheric sciences and physical oceanography (CASPO) department with George Sugihara. After becoming a staff scientist at the Salk institute, he continued work on nonlinear dynamics mainly on causal inference in systems neuroscience and systems biology. He was also a visiting scientist at the National Institute of Industrial Science and Technology (AIST) of the Japanese Ministry of Economy Trade and Industry (METI) in the Information Technology Reseach Institute (ITRI) and the Artificial Intelligence Research Center (AIRC) to make the computational methods in empiric dynamic modeling suitable for Big Data and high performance computing (HPC) using Japan’s second fastest supercomputer ABCI. In addition to this work Dr. Pao identified, cloned and developed cephalopod reflectin proteins from various squid species for the manipulation of the optical refractive index in vivo in mammals. Before joining OIST he had a two year stint in industry as a research director for high throughput screening data science and gene therapy at Vertex pharmaceuticals, a Fortune 500 company.
Abstract: Quantitative science has been dominated by physics that tries to determine the relationships of natural variables in the form of equations. For these to have close form analytical solutions ideally requires that the relationships among variables to be linear, decomposable and the noise hopefully Gaussian. For this reason for highly nonlinear systems Physics type approaches have so far not given good solutions. In the last decade it has become apparent that Deep Learning is able to give surprisingly good mappings for these types of nonlinear problems where traditional physics has not. The drawback of deep learning mediated fitting is that the solutions it provides are “black box” which lack explainability. Neural activity is indeed highly nonlinear. To solve this type of problems in an explainable manner we offer a framework based on the generalized Takens theorem from dynamical systems theory, to generate data driven embeddings of time series on low dimensional manifolds that generically map neural activity to behavior. These embeddings allow prediction of neural activity and behavior, inference of cause and effect relationships of hundreds of thousands of neurons that scales linearly with computational complexity dimensionality, simulate neural activity with associated behavior and guarantee explainability, and experimental testability of inferred relationships. Thus the approach allows falsifiability of the computationally obtained hypotheses. We show a handful of use cases from systems neuroscience of animals and humans where we generate realistic simulations of brains and their corresponding output behavior that are equivalent to single behavior brain downloads.
This seminar is part of Foresight's Neurotech Seminar Series. To join future seminars in this program please apply here.
A group of neuroscience researchers, entrepreneurs, and allies advancing beneficial short-term and long-term neurotechnology applications.
Nominate a seminar presenter/topic
Feel free to reach out to lydia@foresight.org with any questions.