Our understanding of biological systems and their neurological principles has strongly influenced the engineering development of machine learning and artificial intelligence. Yet, computational learning mechanisms only vaguely mimic those found in nature, despite our far greater understanding of the neurosciences and related disciplines than we used to have only a few decades ago. We aim to catalyze this discussion and explore recent developments, challenges, and untapped potential of integrating biological principles with developing the next generation of machine learning and artificial intelligence systems.
Registration
Click below to register for the workshop. (Limited slots available)
Title: Application of AI/ML Approaches in Decoding Mesoscale Brain Networks
The brain is a complex system of optimized interconnected networks across multiple spatiotemporal scales. Among these, mesoscale networks are defined as the intermediate level between individual neural cells to neural ensembles and circuits, playing pivotal roles in signaling and information processing that underpin brain function. Decoding behavior, such as skilled movements, from mesoscale brain networks remains a central goal in neuroscience. In recent years, artificial intelligence (AI) and machine learning (ML) have played an increasingly significant role in elucidating the neural mechanisms underlying motor function. Yet, the ability to simultaneously monitor and decode real-time activity from these networks has been limited by technological constraints. In this talk, I will present AI/ML-based approaches for decoding skilled motor behavior from mesoscale brain networks using data obtained through advanced in vivo two-photon calcium imaging. Furthermore, I will discuss how insights derived from deep learning models can advance brain-inspired neural decoding technologies, offering new avenues for understanding neuronal computation and motor behavior. Personal Websitehttps://medicine.hsc.wvu.edu/neuroscience/faculty-labs/shahrzad-latifi-phd/
G. Bard Ermentrout
Title: Gut Feelings: A data driven approach to modeling the dynamics of the proximal colon
The classic view of the colon is based on the physiology of the distal colon which propels fecal pellets by generating smooth muscle contraction orally and relaxation aborally. This behavior requires the presence of the pellet to activate the sensory motor program. In contrast, no pellets have yet formed in the proximal colon. Furthermore, in absence of any fecal contents, the proximal colon produces regular rhythmic propagating contractions with a period of 2-5 minutes in the mouse. We present new experimental results that shed light on the possible mechanisms for these contractive waves. We use the new experiments to suggest a computational model of the proximal colon. We show that the combination of the circular muscle, sensory neurons unique to the proximal colon, and specialized pacemaker cells, create a double chain of oscillators that is able to reproduce the observed dynamical behavior. By breaking the model into its component parts, we are able to provide a mathematical explanation for the slow rhythmic contractions. We use the model to explain pathological motility patterns in experimentally induced inflammation. Personal Websitehttps://www.mathematics.pitt.edu/people/g-bard-ermentrout
William Nourse
Title: Structure and Function in the Nervous System: Striking the Balance Between Bioinspiration and Biomimicry
Multiple fields, including NeuroAI, BioML, Neuromorphic Computing, and Neurorobotics, aim to extract principles from the brain and nervous system in order to improve the capabilities of artificial systems. While all of these fields have experienced tremendous advances in recent years, they all still grapple with the same core dilemma: how much detail of how the nervous system functions is actually important? In this seminar, we will explore different aspects of this debate in theory and practice, and will present factors to consider when drawing on neural systems for inspiration. Personal Websitehttps://williamnourse.github.io/
Valeriya Gritsenko
Title: AI-Powered Biomechanics: Revolutionizing Motion Analysis for Health and Performance Applications
I will talk about three projects that integrate artificial intelligence with biomechanical modeling to analyze and interpret human motion in challenging or resource-constrained environments. They emphasize scalable, efficient alternatives to traditional motion capture by leveraging neural networks or consumer-grade tools to extract meaningful health and performance metrics. The first project focuses on real-time neural networks performing inverse kinematics for joint angle estimation. The second project evaluates the feasibility of using smartphone and machine learning for evaluating infant head control. The third project will discuss challenges and opportunities in estimating muscle moments from video for astronaut health monitoring. Personal Websitehttps://sites.google.com/site/valeriyagritsenko/Home
Gary Marsat
Title: Building a cerebellar predictive circuit for adaptive filtering
