[4C2-J-1] Fundamental AI, theory: brain-based design of intelligence

[4C2-J-1-01] Structure Estimation of Neural Networks Based on Sparse Modeling

〇Ren Masahiro¹, Toshiaki Omori¹ (1. Kobe University)

Keywords:Neurocomputing, Sparse Modeling, Neural Networks, Data-driven Approach

In recent years, great advances have been made in measurement technology of neuroscience. Particularly, imaging techniques such as voltage imaging and calcium imaging enable us to observe membrane potentials from multiple neurons simultaneously. In order to deepen understandings of brain system, it is crucial to make progress in a data-driven method for extracting neural network dynamics which produce observable data. In this research, we propose an algorithm to estimate neural network structure which is thought to be deeply related to brain system. Using SMC (Sequential Monte Carlo method) and Group LASSO (Least Absolute Shrinkage and Selection Operator), we achieve structure estimation of neural networks. By reconstructing membrane potentials from the estimated neural network structure, we show effectiveness of the proposed method.