SEMINARIO: Connectomic motifs in neural networks

Abstract:
Three key elements for NeuroAI will be presented: objectives, architecture and learning method.
We will show that different connectomes (C.Elegans, Drosophila, Mouse) include neural motifs that may play a role in learning, in particular signed XOR motifs.
Basic examples of computational autoencoders employing bio-plausible methods, using such motifs, will be presented, remarking the role of a latent space.
Finally, we will discuss about the different way knowledge is managed in a brain and in current LLMs, and how our ENGRAMMER project addresses this divergence.