Representation Learning for Reinforcement learning
Representation learning is a subset of machine learning that focuses on learning feature representations from raw data rather than manually designing them. The goal of representation learning is to automatically discover meaningful patterns or features in data that can be used to improve the performance of a machine learning model on a specific task.In traditional machine learning, features are often hand-crafted by domain experts or engineers who have a deep understanding of the problem at hand. However, this approach can be time-consuming, costly, and may not always lead to the best results, especially when dealing with complex data such as images, audio, or natural language.Representation learning, on the other hand, allows the machine learning system to learn these features directly from the data, thereby automating the feature engineering process. This can lead to significant improvements in performance, as the machine learning model can leverage more complex and diverse representations of the data. In recent years, representation learning has become an increasingly important area of research in the field of artificial intelligence. It has been used to improve the performance of a wide range of applications, including computer vision, natural language processing, speech recognition, and recommendation systems. The objective of this thesis is to apply these techniques in the field of Reinforcement Learning, where almost all the currently used methods rely on a "temporal difference" approach. On the contrary, we will use representation learning to solve classical Reinforcement Learning problems with a different path, that is by learning a a mapping from an RL problem to another one in a given family that is more easily solvable.
Back to Current Students