Daphne Chen
@Daphne__Chen
Followers
51
Following
5
Media
1
Statuses
6
PhD student @uwcse, previously @CMU_Robotics & @GeorgiaTech
Joined November 2024
Excited to share our paper on human-AI cooperation using generative models, and the first work of my PhD @uwcse - to be shared at #NeurIPS2024! Thanks to my amazing collaborators @liangyanchenggg @abhishekunique7 @SimonShaoleiDu @natashajaques for making this such a fun project!
π Excited to release our #NeurIPS2024 paper on zero-shot human-AI cooperation. For the first time, we use generative models to sample infinite human-like training partners to train a Cooperator agent. π₯Experience it! πCheck out our π₯π’π―π πππ¦π¨ π
1
4
15
Look ma I'm in this pic! Happy to have gotten first runner up for the Madrona Prize for VAMOS!
Each year, weβre inspired by the creativity and impact of @UW researchers advancing the frontiers of computing. For nearly 20 years, Madrona has partnered with the @uwcse to honor that spirit of innovation through the Madrona Prize, recognizing research that blends technical
1
1
21
Inspired by an annual tradition at my old lab @GTrobotics ππ€
0
0
2
Had a fun time handing out candy all day to trick-or-treaters in the @uwcse lobby using our Stretch robot @hellorobotinc! Happy Halloween from @uwrobotics brought to you by the Human-Centered Robotics Lab @UW :)
2
2
10
Training cooperative AI agents is tricky - most methods miss the diverse behaviors needed for real human-AI coordination. Our new paper introduces GOAT: using generative models + regret-based adversarial training to create realistic yet challenging training partners. π
1/6 Current AI agent training methods fail to capture diverse behaviors needed for human-AI cooperation. GOAT (Generative Online Adversarial Training) uses online adversarial training to explore a pre-trained generative model's latent space to generate realistic yet challenging
1
0
0
Be sure to check out our work in action by playing Overcooked with our AI agents in our live demo π€π§βπ³:
sites.google.com
TLDR: We use generative models to sample infinite human-like partner agents to train a coordinator agent. These agents cooperate well with real human players, achieving better performance compared to...
0
0
2