Explore the world of reinforcement learning with the guide of a distinguished lecturer!
Basic information
- Date & Time: 5th/March 13:30-17:00 6th/March 9:00-13:00
- Location: 1st Science bldg. (理学部1号館) in Hongo Campus
- Room233
- How to Access
- Registration is required!!! (jump to registration form)
Schedule & outline of lecture
We are going to have 4 sessions, 90min each. two of which are held in the afternoon of 5th, and the others are in the morning of 6th. The following is the tentative schedule:
- Session1: 13:30-15:00, 5th/March
- Session2: 15:30-17:00, 5th/March
- Session3: 9:00-10:30, 6th/March
- Session4: 11:00-12:30, 6th/March
Depending on the background of the audience, the lecturer will change the pace and the topic of the lectures. The following topics are going to be touched in the lectures:
- Reinforcement Learning. Agents, environments and rewards.
- The challenges of decision-making: prior knowledge, acquisition of information, learning.
- Markov Decision Processes and the omniscient decision-maker
- Partial observability. Controlling Hidden Markov models with Bayesian updating. The role of models.
- Learning to make good decisions without prior knowledge. Temporal difference learning.
- Multi-armed bandits: "the hydrogen atom" of reinforcement learning.
- Reinforcement Learning in desperately large environments. (if time permits)
Lecturer: Dr. Antonio Celani
Dr. Antonio Celani has received his PhD in statistical physics of non-equilibrium systems. He started his early career working on fluid dynamics and turbulence and achieved substantial contributions to this area (1, 2,3).
After moving to Institut Pasteur, he shifted his research target to biological systems and proposed an innovative method by which we can infer the response characteristics of bacterial chemotaxis only from the tracked trajectories of free-swimming bacteria (4, 5). Concurrently, he also contributed to the development of statistical and stochastic thermodynamics (6).
He has now extended his research activities to odour sensing, searching, and communication by higher organisms like a fly and worked on the problem how biological agents like insects can sense and search the odour, which are conveyed by complicated turbulent flow in the environment (7). His recent work included an application of reinforcement learning to complex and intelligent behaviors of biological agents (8, 9, 10,11) as well as a proposal of an algorithm to balance exploration and exploitation in reinforcement learning by using infomax principle (12).
He is definitely one of the leading scientists who challenge the problem of biological intelligence by integrating the knowledge from three different regimes: transfer of chemical signals by physical processes like turbulence and diffusion (fluid dynamics); reception of the signal by stochastic biochemical processes (stochastic and statistical thermodynamics); decision making by biological agents based on the histories of the received signals (information processing and learning)
Can't you wait till March?
You can watch much shorter version of his lecture here.