Notions of intrinsic motivation



Consciousness Club Tokyo

http://conscious-machine.org/club/

Speaker: Martin Biehl (Araya, Inc.)
Title: Notions of intrinsic motivation
Venue: Araya, Inc., Tokyo, Japan (http://www.araya.org/eng)
Date and Time: Oct. 19, 2017, 7-9 PM JST

Abstract:
Intrinsic motivation is gaining attention in AI and machine learning. This is mainly due to machine learning papers successfully employing intrinsic motivations to solve tasks in reinforcement learning tasks with sparse rewards.
Originally the notion goes back to developmental psychology and was picked up by computer scientists starting in the 1990s. One strand of this research, called developmental robotics, is inspired by the way human children acquire an understanding of their capabilities (from simple movements to walking, self-recognition, and even high level skills such as language). In contrast to its use in machine learning algorithms the intrinsic motivation here is usually the only objective function (reward function) and not just an additional term. In this context an intrinsic motivation is also expected to drive continual, open ended learning. This in turn can be seen as an approach to building artificial general intelligence.
In this talk I will try to give an overview over existing formulations of intrinsic motivations and discuss the different use cases.


Post time: Oct-19-2017
INQUIRY NOW
  • * CAPTCHA: Please select the Tree