[Seminar] Professor Hiroshi Makino: Learning in intelligent systems

We are excited to have an online seminar by Prof. Hiroshi Makino from Nanyang Technological University in Singapore. His lab combines large-scale neural recordings of mouse cortical neurons during a sensorimotor task with deep reinforcement learning algorithms to discover commonalities and differences between animal and machine intelligence (Suhaimi et al. Science Advances. 2022). Everyone is welcome to join the seminar!

Learning in intelligent systems

Abstract:

The overarching goal of brain science is to explain various domains of natural intelligence by constructing accurate theoretical models under the biological constraints. Despite the recent progress, however, current models heavily depend on supervised and unsupervised learning, which themselves may not simply translate into candidate models of how the brain learns to improve its behavior. I argue that deep reinforcement learning (RL), integration of deep neural networks and theories of RL, generates hypotheses on natural intelligence underlying reward-based learning and decision-making. By minimizing the difference in behaviors and representations between artificial agents and the ground truth of biological agents in a supervised manner, deep RL models can be iteratively optimized while alternative hypotheses are rejected. Such reverse-engineering of the brain via direct comparisons between the artificial and biological system provides normative accounts for how natural intelligence is implemented via concerted actions of neurons. Here I will describe our recent effort to demonstrate the strength of our approach to understand learning in intelligent systems.

You can join the seminar via ZOOM (meeting ID: 782 721 4941, Password: 436475).