Kevin Max: Learning, forward and backward: Error backpropagation in cortical circuits
| When |
Oct 16, 2025
from 02:00 PM to 03:00 PM |
|---|---|
| Where | Bernstein Center, Hansastr. 9a, Lecture Hall. |
| Contact Name | Gundel Jaeger |
| Add event to calendar |
 vCal iCal
|
Abstract
How does cortex learn to perform complex functions? Classical predictive coding models (Rao & Ballard, 1999) propose that learning is based on mismatches between expected and actual stimuli. Similarly, the error backpropagation algorithm, underpinning the deep learning revolution, enables efficient learning through neuron-specific error signals. This motivates theories of efficient learning in cortex based on the propagation of mismatch/error signals.
However, such bio-plausible models of learning in cortex typically operate on high levels of abstraction. Common issues include: oversimplified neuron models, unexplained information sharing between distant synapses and neurons, biologically implausible network architectures, and phased (interrupted) learning.
In this talk, I will discuss a biologically motivated, multi-area circuit model of cortex. I will show that such networks can learn complex tasks using multiple cortical areas, while systematically addressing the aforementioned issues of bio-plausibility. Finally, I will discuss ongoing experiments to uncover how cortex learns from errors.