Suggestions
Natasha Jaques
Incoming faculty at University of Washington, Senior Research Scientist at Google
Natasha Jaques is a professional known for her expertise in Social Reinforcement Learning, utilizing multi-agent RL techniques to enhance single-agent learning, generalization, and human-AI interaction.
With a strong background in deep learning, reinforcement learning, and machine learning, Natasha Jaques has made significant contributions to the field.
She pursued her education at several esteemed institutions, including a PhD from the Massachusetts Institute of Technology, a Master of Science in Computer Science from The University of British Columbia, and a Bachelor of Science Honours in Computer Science and Bachelor of Arts in Psychology from the University of Regina.
Natasha Jaques has held various roles in prestigious organizations such as being an Assistant Professor at the University of Washington, a Senior Research Scientist at Google, and a Research Intern at DeepMind and Google Brain.
Her rich professional history includes being a Graduate Teaching Assistant at Massachusetts Institute of Technology, a Brain Intern at Google, a PhD Candidate at MIT Media Lab, and a Program Manager Intern at Microsoft.
In addition, she has also served as Girlsmarts Coordinator and Graduate Teaching Assistant at The University of British Columbia, Sessional Lecturer, Research Assistant, and Supplemental Instruction Leader at the University of Regina, and as an Animatrice at Trait d’Union Outaouais Inc.
For further information and updated details, visit Natasha Jaques's website at https://natashajaques.ai/. You can explore her publications on Google Scholar at https://scholar.google.com/citations?user=8iCb2TwAAAAJ&hl=en.
Highlights
Excited to be speaking at the Montreal IVADO Bootcamp dedicated to Autonomous Agents, where I will give a remote talk on Multi-agent Reinforcement Learning for LLMs on August 15th at 11am ET.
Full schedule and registration: https://t.co/28Hue8zKK7
In our latest paper, we discovered a surprising result: training LLMs with self-play reinforcement learning on zero-sum games (like poker) significantly improves performance on math and reasoning benchmarks, zero-shot. Whaaat?
How does this work? We analyze the results and find that LLMs learn emergent reasoning patterns like case-by-case analysis and expected value calculation that transfer to improve performance on math questions.
This work shows the benefit of RL training for improving reasoning skills when there is no possibility for data leakage. AND how continuously evolving multi-agent competition leads to the development of emergent skills that generalize to novel tasks. Read more below!