Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Biophysical Deep Learning: Laminar-resolved Cortical Columns as Neural Network Units

Dasja de Leeuw¹, Rainer Goebel, Mario Senden¹; ¹Maastricht University

Presenter: Dasja de Leeuw

Cognitive computational neuroscience strives to develop models that achieve both biological and cognitive fidelity. Propelled by the success of deep neural networks (DNNs) in emulating human functional capacities and neural representations, the field increasingly utilizes DNNs for generating, testing, and refining theories of the neurocomputational processes. However, these neuroconnectionist models often lack neuroanatomical detail and neuronal population dynamics, factors that provide important constraints on the neurocomputational solution space. To address this, we introduce a biophysics-informed neuroconnectionist modeling approach with powerful learning capabilities. Our approach constructs neural networks from laminar-resolved cortical columns with neuroanatomically-realistic internal connectivity. In a proof of concept, we show that these networks can be succesfully trained to reproduce firing rates of neuronal populations in a perceptual decision-making task and achieve high accuracy in classification tasks. This work demonstrates the feasibility of embedding function in biophysics-informed models and introduces a new class of neuroconnectionist models striking a meaningful balance between biological realism and cognitive function.

Topic Area: Brain Networks & Neural Dynamics

Extended Abstract: Full Text PDF