چكيده به لاتين
Transfer learning is a particular category of machine learning methods that extract knowledge from one task and transfer it to another task. Due to that issues that are solvable using machine learning methods are getting harder and larger each and every days And, consequently, the longer the training time of learning methods requires the need for methods that can use prior knowledge of humans or other learning methods.
In the transfer learning literature, the proposed methods can be divided into two general groups. Methods that transfer knowledge, and methods that extract knowledge.In the literature, transfer methods are well discussed and different methods are presented. But knowledge extraction methods are less explored. While the type of knowledge we transfer can be very important and lead to improvement.
Among the knowledge extraction literature, the subject that has never been addressed is learning process and no knowledge extraction method has been provided for learning process, while the experiences of the agent during the learning process can be very helpful and help improve the learning process. Therefore, we decided to present in a methode that extracts knowledge from learning process that can be used to improve the learning speed.
In this thesis, we have presented a potential function that attempted to extract knowledge from the learning process and, using this knowledge, increased the learning speed of single-task and multi-task agents. The proposed method has been evaluated in the Arcade learning environment and the results indicate an improvement in the learning process in both single-task and multi-task agents.
