Conversations with data: Advancing the state of the art in language-driven data exploration
One key aspiration of AI is to develop natural and effective task-oriented conversational systems. Task-oriented conversational systems use a natural language interface to collaborate with and support people in accomplishing specific goals and activities. They…
Strategically Efficient RL
This project implements exploration mechanisms for reinforcement learning that are specifically designed for non-cooperative multi-agent settings. Exploration mechanisms for single-agent RL (e.g. curiosity) are ill-suited to non-cooperative multi-agent settings, as they fail to account for…
Synthesized Collective Communication Library (SCCL)
The Synthesized Collective Communication Library is a tool for synthesizing collective algorithms tailored to a particular hardware topology. This project creates high-performance collective communication algorithms for the uncommon interconnect topologies connecting AI accelerators inside servers.…
Glider
We release Glider, a reinforcement learning agent to extract UI scripts from the web. Given a website URL (e.g., http://www.unitconversion.org/ (opens in new tab)) and a task description in natural language (e.g., “convert length 7333…