Electrical and Computer Engineering Faculty Candidate Seminar
Rainer Engelken
Postdoctoral Researcher, Columbia University
Wednesday, March 26, 2025, 10:00-11:00 am
B02 CSL Auditorium or Online via Zoom
Title: From Brains to Machines: Engineering Efficient, Reliable Neural Computation using Dynamical Systems Theory
Abstract: Modern artificial neural networks achieve impressive performance but require vast computational resources, while biological brains operate with remarkable efficiency and robustness. This talk presents a dynamical systems approach to neural computation, providing theoretical insights that improve stability, scalability, and efficiency in both artificial and biologically inspired networks. First, I will demonstrate how controlling network stability by dynamically adjusting Lyapunov exponents mitigates exploding and vanishing gradients, greatly improving learning in both biological and artificial neural networks. Next, I will focus on spiking networks, promising candidates for energy-efficient neuromorphic computing due to their brain-like operation. I show how to reduce computational costs of simulating spiking networks from O(N) to O(log N), where N is the number of neurons, using a novel event-based simulation algorithm. This enables efficient, large-scale simulations for neuromorphic computing and neural circuit modeling. Then, I will show how biophysical properties of single neurons influence network dynamics, revealing a regime where spiking networks intrinsically suppress chaos, enhancing both reliability and information transmission. By bridging engineering, physics, and neuroscience, these insights have direct implications for designing more scalable, robust, and efficient AI systems and developing brain-like models in theoretical neuroscience and neuromorphic computing.
Rainer Engelken is a postdoctoral researcher at Columbia University's Center for Theoretical Neuroscience in Larry Abbott’s lab. His research bridges dynamical systems theory, neuroscience, and AI, focusing on neural computation, learning, and efficiency in artificial and biological networks. He earned his Ph.D. at the Max Planck Institute for Dynamics and Self-Organization (Göttingen, Germany), where he studied chaotic dynamics in spiking neural networks.