Offline Reinforcement Learning (Offline RL) is the problem where an RL agent aims to learn a good policy from a fixed dataset of pre-collected experience of interaction with an environment. Due to its great potential in making RL applicable in real-world applications where interaction with the environment is costly and/or risky, offline RL has recently gained attention from the RL research community, and several offline RL algorithms have been developed that try to overcome the inefficiency of directly using traditional online RL algorithms for offline RL.

In this project, we consider the offline multi-agent RL (offline MARL) problem which extends offline RL to more than one agent, and, in particular, we focus on the cooperative setting where the agents aim to accomplish a shared task. The core research question of this project is: are the existing offline RL algorithms still effective for offline cooperative MARL? To answer this question, we plan to 

1) select a subset of existing offline RL algorithms that can be relatively easy to adapt to the multi-agent setting, and 

2) implement the adapted algorithms to evaluate their effectiveness. We expect that aspects of the multi-agency (e.gexponentially growing action spaces) could bring about additional challenges for offline RL, such that many, if not all, adapted offline algorithms will not be effective.

• 2/16/23 Katrina Pitts: Charting Your Path to Achievement (5:00 PM | Starks-L) (Slides|Recording|Feedback)

• 2/22/23 Junsoo Lee:  Research Outside of Your Boundary (5:00 PM | N.A. Jenkins Board Room) (Slides|Recording|Feedback)

• 3/1/23 Bharat Rawal: Exploratory Research Steps: How to convert your question into exploratory research in the computing field according to IEEE Standard (5:00 PM | Student Lounge - Swinton Campus Center) (Slides|Recording|Feedback)

• 3/15/23 Brittany Morrone & Cornell Sneed: How to Create a Winning Team (5:00 PM | Student Lounge - Swinton Campus Center) (Slides|Recording|Feedback)

• 3/22/23 Biplav Srivastava: Open Social Innovation (5:00 PM | Student Lounge - Swinton Campus Center) (Slides|Recording|Feedback)

• 3/30/23 Adrien Ratsimbaharison: Computer-Assisted Qualitative Research (5:00 PM | Student Lounge - Swinton Campus Center) (Slides|Recording|Feedback)

• 4/5/23 Andrew Sabalowsky: Using Matrices for Complex, Multivariate Decision-making (5:00 PM | Student Lounge - Swinton Campus Center) (Slides|Recording|Feedback)

• 4/12,19/23 Students' service-learning projects presentation (TBD)

• 4/21/23 Discover USC 2023 (Columbia Metropolitan Convention Center) Sign-up

We are now accepting proposals from BS and UofSC faculty members and professionals to host workshop sessions (stipend provided!). Please check our Call for Proposals.

Workshop topics include but are not limited to:

• Build up students’ self-efficacy in computing research.

• Guide students to analyze and know their potentials; motivate them to explore their vision, passion and career goals.

• Tips and strategies for efficient literature review, research project design and planning.

• Team bonding, effective communication and collaboration skills on a group project.

• Project timeline setting, progress tracking, and project management.

• Project assessment and evaluation. 

Who can submit: 

• Experienced faculty or professionals in related areas, such as psychology, education, project management, computer science, etc.

Where to submit: 

• Please submit your proposal via this link:  https://forms.office.com/r/SQbucDnGbS 

Time commitment: 

• We will have bi-weekly meetings to share and discuss information about the workshop.  

Stipend:

• $500 for developing a workshop session (2022 fall semester).

• $500 as a speaker honorarium (2023 spring semester).

Submission Deadline:

• Nov 4, 2022 

Notification:

• Nov 10, 2022