Project

How to handle data shift during fine tuning RL models?

Positions: Student Researcher

Created: 2023-10-22 Deadline:

Location: Poland

Foundation models have delivered impressive outcomes in areas like computer vision and language processing, but not as much in reinforcement learning. It has been demonstrated that fine-tuning on compositional tasks, where certain aspects of the environment may only be revealed after extensive training, is susceptible to catastrophic forgetting. In this situation, a pre-trained model may lose valuable knowledge before encountering parts of the state space that it can handle. The goal of the project is to research and develop methods which could prevent forgetting of the pretrained weights and therefore get better performace by leveraging previous knowledge. Highly recommend section 4.4 Minecraft RL

General must-have requirements

PyTorch

Contact: Maciej Wołczyk, Bartłomiej Cupiał (maciej.wolczyk [ at ] gmail.com)

Project's lab:

Group of Machine Learning Research

Location: Kraków, Poland Website: gmum.net Available projects: 35

GMUM (Group of Machine Learning Research) is a group of researchers working on various aspects of machine learning, and in particular deep learning - in both fundamental and applied settings. The group is led by prof. Jacek Tabor. We are based in the Jagiellonian University in the beautiful city of Kraków, Poland.

Some of the research directions our group pursues include:

• Generative models: efficient training and sampling; inpainting; super-resolution,
• Theoretical understanding of deep learning and optimization,
• Natural language processing,
• Drug design and cheminformatics,
• Unsupervised learning and clustering,
• Computer vision and medical image analysis.

In 2023, we organized the second edition of Machine Learning Summer School (MLSS^S) with a focus on Applications in Science. We invite participants to collaborate with us on various ongoing research projects - learn more here.

See lab's page