Modeling natural neural networks of decision making with artificial neural networks.

One main focus in neuroscience is to understand the relationship between decision making and various brain regions. Researchers use machine learning approaches to model the neural circuits of cerebral cortices, cerebellum, and basal ganglia. This review focuses on artificial neural networks (ANNs), particularly recurrent neural networks (RNNs), to model cortical functions for decision making. We first introduce the basic architecture of RNNs and explain how researchers compare the activity and circuits between artificial and biological networks. We also summarize how RNNs model the prefrontal and posterior parietal cortical in tasks involving short-term memory, perceptual decision making, and value-based decision making. We then show our recent challenges to develop a real-cyber hybrid network, that integrates neuronal activity in mice with RNN-based artificial units to better generate continuous-time body movements, compared to conventional RNNs that only use artificial units. The hybrid network tries to develop RNNs which have similar activity to the brain by using real neurons, rather than developing artificial RNNs and comparing their functions with biological brain. We propose that such integrative approaches in neuroscience and AI will further our understanding of both natural and artificial intelligence in the field of neuro-AI.