Abstraction Networks
Lead Research Organisation:
Imperial College London
Department Name: Computing
Abstract
A Method For Knowledge Extraction Towards Lifelong
Reinforcement Learning
. Due to the requirement of autonomy, this project examines progress made in Lifelong
Reinforcement Learning (LRL) and identifies a promising development direction in the use
of technologies which incorporate a relational inductive bias over the knowledge representations. Finally, by integrating two prominent methods, namely Graph Networks and Latent
Feature Models, Abstraction Networks are proposed as a potential LRL framework, as they
assume a network of fundamental state-abstractions that have inherent synergy with the
knowledge manipulation requirements of lifelong learning.
Research area Reinforcement learning
Reinforcement Learning
. Due to the requirement of autonomy, this project examines progress made in Lifelong
Reinforcement Learning (LRL) and identifies a promising development direction in the use
of technologies which incorporate a relational inductive bias over the knowledge representations. Finally, by integrating two prominent methods, namely Graph Networks and Latent
Feature Models, Abstraction Networks are proposed as a potential LRL framework, as they
assume a network of fundamental state-abstractions that have inherent synergy with the
knowledge manipulation requirements of lifelong learning.
Research area Reinforcement learning