اخبار و رویدادها

Skill based transfer learning in continuous reinforcemnet learning domain

Skill based transfer learning in continuous reinforcemnet learning domain


Agents, physical and virtual entities that interact with their environment, are becoming increasingly prevalent. However, if agents are to behave intelligently in complex, dynamic, and noisy environments, we believe that they must be able to learn and adapt. The reinforcement learning (RL) paradigm is a popular way for such agents to learn from experience with minimal feedback. Indeed, RL allows autonomous agents to learn to improve their performance with experience in an unknown environment. The required learning time and the curse of dimensionality restrict applicability of Reinforcement Learning (RL) on real robots. A vast number of RL studies shows that it is believed that the curse of dimensionality can be lessen, to a great extent, by implementation of state abstraction methods and hierarchical architectures. While significant progress has been made to improve learning in a single task, the idea of transfer learning has only recently been applied to reinforcement learning tasks. The insight behind transfer learning (TL) is that generalization may occur not only within tasks, but also across tasks. So, Transfer learning has recently gained popularity due to the development of algorithms that can successfully generalize information across multiple tasks.

The idea of transfer of knowledge in order to improve the performance of machine learning algorithms stems from psychology and cognitive science research. A vast number of psychological studies show how the effectiveness of learning a task is strictly related to the knowledge retained from solving similar tasks. For instance, a person who can drive a bicycle learns to drive a motorcycle faster than a person who has never driven anything similar. The reason is that, while learning how to drive a bicycle, the human mind retains abstract knowledge about the problem of driving that can be profitably reused when facing a problem that shares some characteristics with driving a bicycle. Human beings can learn amazingly fast because they effectively bias the learning process towards a very limited set of solutions obtained by transferring the knowledge retained from solving similar tasks. Similarly, the idea of transfer learning is that it is possible to improve the performance of machine learning algorithms by biasing their hypothesis space towards a set of "good" hypotheses according to the knowledge retained from solving other tasks.

Here, we aim to use transfer learning techniques in order to make reinforcement learning algorithms more efficient by incorporating knowledge from previous tasks.

 

 

Provider

فرزانه شعله
email: f.shoeleh@ut.ac.ir

 
 

Supervisor

مسعود اسدپور
email: asadpour [AT] ut.ac.ir

 

 
آدرس کوتاه :