Prof. Yukie Nagai


Biography:

Yukie Nagai is a Project Professor at the International Research Center for Neurointelligence, the University of Tokyo. She received her Ph.D. in Engineering from Osaka University in 2004 and worked at the National Institute of Information and Communications Technology, Bielefeld University, and Osaka University. Since 2019, she leads Cognitive Developmental Robotics Lab at the University of Tokyo. Her research interests include cognitive developmental robotics, computational neuroscience, and assistive technologies for developmental disorders. Her research achievements have been widely reported in the media as novel techniques to understand and support human development. She also serves as the research director of JST CREST Cognitive Mirroring.


Title of talk:

Robot intelligence inspired by human intelligence



Abstract:

Human intelligence can provide a new design principle for robot intelligence. Especially, the mechanism of cognitive development has the potential to enable robots to learn to acquire various cognitive functions through their sensorimotor experiences. Open questions are what neural mechanisms underlie cognitive development and what computational models can replicate these mechanisms. My talk presents computational neural networks that model a neuroscience theory called predictive coding. The theory of predictive coding suggests that the human brain works as a predictive machine; that is, the brain does not simply rely on incoming sensory signals but rather predicts the signals using its internal models and perceive the world and act on it by integrating the sensations and predictions. The process of minimizing prediction errors, which are the discrepancy between the sensory signals and predictions, produces both perception and action. We have been investigating how the neural networks based on predictive coding enable robots to acquire cognitive functions and to what extent the networks replicate human-like development. Our first experiment demonstrates that the robots acquire social cognitive abilities such as reading intention and helping others through the minimization of prediction errors. The internal models acquired through sensorimotor experiences replicate the function of mirror neuron systems and enable the robots to infer the internal states of others and establish proto-social behaviors. Further experiments show that altered predictive processing leads to individual diversity in cognitive functions. Aberrant predictive processing results in difficulties in learning and/or adaptation as observed in developmental disorders. I will discuss how our computational approach contributes to the design of human-like robot intelligence as well as to a better understanding of human intelligence.