I recently defended my PhD on how RL agents can learn with and from other agents!
I’m truly grateful to my advisor, Peter Stone, who has shown me how to conduct research with vision, integrity, and rigor, while fostering a great environment for collaboration and growth. I’m also grateful to the many amazing people I’ve encountered along the way, including my mentors, who believed in my ideas while teaching me how to make them better, my labmates, many of whom became valued collaborators and friends, and my mentees, for their fresh perspectives and energy. I’m also grateful to my partner Ben, for his unwavering support along the way.
If you’d told me at the beginning of my PhD that I would be sad when it was over, I wouldn’t have believed you. But I’m indeed sad to move on from LARG, which has become my home.
I’m excited for the next phase of my life, where I’ll be working on multi-agent systems at Google DeepMind London, as a research scientist!
Friends at #AAMAS, tomorrow at 11 AM at the ALA workshop, I will be giving an early presentation of our #ICML2026 Spotlight, "Accelerating Q-learning through Efficient Value-sharing across Actions". Done in collaboration with Brett Daley, @white_martha, and @MarlosCMachado.
Tomorrow (08/08) at @RL_Conference, I will be presenting our (Brett Daley, @MarlosCMachado@white_martha) work:
"An Analysis of Action-Value Temporal-Difference Methods That Learn State Values" in CCIS 1-160 from 11:45-12:30.
Paper: https://t.co/3Z3WiAi95V.
Poster: 43.
The Most Thought-Provoking paper award goes to Thinking is Another Form of Control 🏆 Congratulations to @JosiahHanna and Nicholas E. Corrado! 🎉 Check out their paper here: https://t.co/vFUSSf4LkC
Hi! I am at #ICML2025 . I'll present a poster on Wednesday morning. I'll talk about using bisimulation-based representation learning for accurate and stable offline policy evaluation. Would love to chat with folks here, so please stop by! @JosiahHanna@WisconsinCS
Excited to be at ICML’25!!! I'll present papers on improving LLM inference and evaluation and pseudolabeling-based semi-supervised learning.
Come and say hi during these sessions, or chat anytime during the week!
[C1]. Prune 'n Predict: Optimizing LLM Decision-making with Conformal Prediction
Authors: Harit Vishwakarma, Alan Mishler, Thomas Cook, Nic Dalmasso (@Mr8ND), Natraj Raman, Sumitra Ganesh
Paper: https://t.co/U2IOpX4lqG
Short talk: https://t.co/nZS1fRcmc8
[C2] Rethinking Confidence Scores and Thresholds in Pseudolabeling-based SSL
Authors: Harit Vishwakarma, Yi Chen (@kitkatdafu), Srinath Namburi (@srinath_namburi), Ramya Vinayak (@ramyavinayak), Fred Sala (@fredsala)
Paper: https://t.co/Byl45YlWn4
Short Talk: https://t.co/5kGass77qz
[W1] Time To Impeach LLM-as-a-Judge: Programs are the Future of Evaluation
Authors: Tzu-Heng Huang (@zihengh1), Harit Vishwakarma, Fred Sala (@fredsala)
Paper:
Crisis = Opportunity. I lost my funding due to federal freezes at Harvard.
Seeking roles to build next-generation agentic AI using RL+LLMs.
🔥 DMs open
Expertise: #RL#LLMs#Robotics#GenAI#DataScience
Published in top-tier CS venues.
https://t.co/4ksfgnxymm
Huge congratulations to @PeterStone_TX for receiving the ACM Allen Newell Award! 🎉 We are proud to celebrate this incredible recognition of his groundbreaking contributions to AI and computer science.👏 #ACM#AAAI#AllenNewellAward
🙌 Meet the 2024 ACM Technical Awards Recipients!
We’re proud to honor this year’s innovators in autonomous systems, cryptography, and software for parallel computers:
🏆 Peter Stone – ACM-AAAI Allen Newell Award
For significant contributions to the theory and practice of artificial intelligence (AI). @UTAustin@SonyAI_global
🔗 https://t.co/U4vpUvBgyX
#SonyAI’s paper "A Super-human Vision-based RL Agent for Autonomous Racing in Gran Turismo" wins Outstanding Paper Award at the 2024 RL Conference! https://t.co/jwUc8sHiro 🏆 #ReinforcementLearning#GranTurismo
Working with @PeterStone_TX during my Ph.D. was one of the greatest privileges. I know everyone who has worked with him would agree. Sony AI is fortunate to have him!
Super excited for this workshop! Our panels and talks will cover a diverse set of perspectives on RL, which should make for an exciting and thought-provoking day. Also looking forward to seeing all the ideas people will submit, especially the contrarian ones.
If you’re at AAAI tomorrow, I’ll be giving a new faculty highlight talk describing our recent work toward the safe deployment of reinforcement learning through offline policy evaluation. The talk is in room 111 at 11:30 am.
If you're hoping to build discussion and community around a topic in RL research, @RL_Conference would be an ideal place to hold a workshop! @daniel_s_brown and I are looking forward to seeing your proposal!
Empirically, we find two important results: 1) ROPE is robust to hyperparameter tuning and 2) under settings of extreme distribution where other algorithms diverge, ROPE is able to significantly mitigate divergence!
Hello! Come see our poster on representation learning for more accurate off-policy evaluation (OPE) at #NeurIPS23@NeurIPSConf (joint work w/ @JosiahHanna@WisconsinCS). See details below! Paper: https://t.co/0iLQZq36nL
These ROPE representations are then fed into fitted Q-evaluation (FQE) to evaluate the evaluation policy. ROPE 🪢 is able to estimate the value of the evaluation policy significantly more accurately than vanilla FQE and other OPE representation learning algorithms.