C02 Ryusuke Hayashi

C02 Neural mechanisms of predictive information processing in macaque visual cortex　

In AI and computer vision research, deep learning using neural networks has realized some aspects of human cognitive functions, such as general object recognition. Furthermore, it has been revealed that the information representation within deep neural networks trained for general object recognition is highly similar to the neural representation in the visual cortex of animal brains (Hayashi & Nishimoto, 2013, et al.). On the other hand, according to the information theory about the brain, the visual cortex is regarded to perform predictive coding in response to constantly changing image inputs. However, it remains unverified whether a deep predictive coding model, implemented as a neural network based on the unified theory of the brain, can quantitatively explain large-scale neural activity data simultaneously recorded from multiple brain regions.

In this study, we plan to simultaneously record neural activity from multiple brain regions along the ventral visual pathway, which is responsible for object recognition, in macaque monkeys while they observe images. We aim to examine how feedforward/feedback signals between brain regions change over time by comparing them with the internal representations of a deep neural network trained for image encoding and prediction-generation processes. Additionally, we plan to explore how neural activity changes under dissociative anesthesia, which induces perceptual changes similar to those seen in schizophrenia, by comparing it with computational models from the perspective of predictive information processing abnormalities.