Tom Tetzlaff: Processing very long sequences in biological neuronal networks
| When |
Jun 25, 2025
from 12:15 PM to 01:00 PM |
|---|---|
| Where | Bernstein Center, Hansastr. 9a, Lecture Hall. |
| Contact Name | Gundel Jaeger |
| Add event to calendar |
 vCal iCal
|
Abstract
The data processed by the brain is sequential. Sensory perception, motor activity generation, language comprehension and production, planning, or solving mathematical equations are sequential processes. Learning, predicting, and recalling sequential data, as well as detecting anomalies in such data streams, are hence fundamental computations performed by the brain. A main challenge in processing such data is context dependency: the correct prediction or recall of an upcoming element in a sequence does not only depend on the previous element, but on the entire history. In [1], we devised an algorithm capable of performing these computations based on a recurrent network of spiking neurons with biophysically interpretable variables and parameters ("spiking Temporal Memory" [sTM]). The sTM algorithm learns complex sequences in a continual, unsupervised manner by means of local synaptic plasticity mechanism known from biology. Prediction and recall of sequence elements are represented by dendritic action potentials. The processing of sequential data in sTM networks is based on ultra-sparse spiking activity, and hence highly energy efficient. So far, the sequence processing capabilities of the sTM algorithm have been demonstrated only for small sequence sets containing few tens of elements.
In this talk, I will show that even small sTM networks representing local cortical microcircuits at the sub-millimeter scale can successfully process large sequence sets containing thousands of sequence elements. Mathematical and numerical analysis reveals that the network capacity is primarily limited by the network dynamics in response to unpredicted, ambiguous stimuli. Close to the capacity limit, sTM networks exhibit synapse densities and spiking activity characteristics reminiscent of the neocortex of awake, behaving mammals.
[1] Bouhadjar, Y., Wouters, D. J., Diesmann, M., & Tetzlaff, T. (2022). Sequence learning, prediction, and replay in networks of spiking neurons. PLOS Computational Biology, 18(6), e1010233. doi:10.1371/journal.pcbi.1010233