Online Brain Science Seminar Series (BSS)

Wolfgang Maass, Institute of Theoretical Computer Science at TU Graz

Date/Time

Thursday, 16:00-17:30 , April 14 2022 (JST) (April 14th, 9:00-10:30 CET)

Title / Abstract

Computations in Cortical Microcircuits: Lessons from large-scale brain models

After having worked for many years on more-or-less abstract models for neural networks of the brain, I became curious about the real thing:

Models for cortical microcircuits that integrate most currently available anatomical and neurophysiologcal data, such as the model of (Billeh et al., 2020) for mouse V1 from the Allen Institute.

We first examined the way how this neural network is structured through the genetic code: Through specific distance-dependent connection probabilities between over 100 different neuron types. I will report results from (Stöckl et al., 2021) and more recent work where we aimed at providing answers to two fundamental questions: How can innate computing capabilities arise in the brain although the genetic code does not have enough information capacity to determine individual synaptic weights, and why the brain uses so many different neuron types.

We then analyzed what additional computational capabilities this large-scale model of V1 acquires if one tunes in addition to its genetically encoded structure the values of its synaptic weights. We found that this model becomes then able to multiplex a number of complex visual processing tasks that have also been used in behavioral experiments. I will report results of (Scherr et al., 2021) where analyzed how this model solves the frequently considered image-change detection tasks, in particular the contribution of more complex data-based generalized LIF (GLIF3) neuron models to this computation.

If time permits, I will also quickly report some results from (Chen et al., 2021), where we compared the visual processing and neural coding strategy of this V1 model with that of CNNs. Our results suggest that data-based V1 models may become an interesting alternative to CNNs, especially with regard to robust temporal processing of visual information.

REFERENCES
Billeh, Y. N., Cai, B., Gratiy, S. L., Dai, K., Iyer, R., Gouwens, N. W., … & Arkhipov, A. (2020). Systematic integration of structural and functional data into multi-scale models of mouse primary visual cortex. Neuron, 106(3), 388-403.
Chen, G., Scherr, F., & Maass, W. (2021). Analysis of visual processing capabilities and neural coding strategies of a detailed model for laminar cortical microcircuits in mouse V1. bioRxiv.
Scherr, F., & Maass, W. (2021). Analysis of the computational strategy of a detailed laminar cortical microcircuit model for solving the image-change-detection task. bioRxiv.
Stoeckl, C., Lang, D., & Maass, W. (2021). Probabilistic skeletons endow brain-like neural networks with innate computing capabilities. bioRxiv.

Host: Taro Toyoizumi

Prior registration required (deadline: Noon March 31, 2022)

Those who have already registered for 2021-2022 series⇒　No need to register again.
1st time registrant⇒go to registration.

Registration

For inquiries

riken-cbs-hd[at]ml.riken.jp