Bio-inspired AI and "AI-inspired biology"
Transfer of knowledge from basic research in neuroscience to applied research in machine learning and electronics is not a one-way street. While biology has a long history of serving as an inspiration for technology and algorithms, a steadily increasing body of work adopts synthetic computational paradigms to develop models for biological cognition. In this talk, I will highlight some of our recent work that addresses both pathways of inspiration.
First, I will discuss how abstract models of prediction (in time and space) and creativity could guide our understanding of cortical computation. In particular, I will show how incorporating inspiration from biology - such as prospective neuronal activity or bizarre dreams during REM sleep - can aid deep learning and memory consolidation. In this context, I will address the role of dendrites, which come with computational advantages and challenges of their own.
In the second part of the talk, I will move to codes that rely on the (relative) timing of spikes. First, I will discuss a deterministic model of deep learning using spike-latency coding, including an implementation on high-speed, low-power neuromorphic hardware for machine vision. Second, I will assume a Bayesian point of view, and present recent results in sampling-based computation with spiking neurons, ending with a rather unusual application of neuromorphic hardware far outside the realm of biology.
Back to the Schedule