CMU Foundation and Language Model Seminar
— 12:30pm
Location:
In Person and Virtual - ET
-
ASA Conference Room, Gates Hillman 6115
Speaker:
KATERINA FRAGKIADAKI
,
JPMorgan Chase Associate Professor of Computer Science, Machine Learning Department, Carnegie Mellon University
https://www.cs.cmu.edu/~katef/
This talk presents our work on LLMs for robot learning, such as manipulation and autonomous driving. LLMs propose tasks, decompose them over time and predict python reward functions, which we then train RL policies for, as seen in https://gen2sim.github.io/. Similarly, using memory-augmented LLMs for embodied agents allows for learning during deployment through memory expansion, highlighted in https://helper-agent-llm.github.io/. Additionally, we discuss driving in the hardest scenarios with a teacher sitting next to you, where the LLM shapes the reward function optimized through a newly proposed optimization method.
Pizza Lunch Provided.
In Person and Zoom Participation. See announcement.
Event Website:
https://cmuflame.org/events.html