| Abstract [eng] |
Reinforcement learning is a very rapidly growing field of research, where a lot of new innovations, state-of-the-art methods and technics are emerging because of never ending challenges for the artificial intelligence. RL gives an opportunity for researchers to investigate how an agent can learn a specific task by simply interacting with the environment. Step by step, after each positive and negative feedback, after millions or trillions of trials and errors in a closed environment, as a result the agent becomes an expert in the field of training. Consequently, the agent is capable to outperform human mind in a restricted domain. From this follows a problem about context switch. After learning to complete one task the agent needs to be retrained from the start again on the new environment. But is there a way to make an agent generalize the knowledge from previous challenges? To answer the question, the RL problem is transitioned to sequence analysis problem. Thesis emphasizes the problem and revolves around the reinforcement learning, transferring and deep learning of sequence analysis. |
