Hritik Bansal

Finally defended my Ph.D. thesis! 🥳 A very warm thank you to my family, friends, and advisors — @kaiwei_chang, @adityagrover_, @VioletNPeng, and Hongjing Lu. Next, I will be joining @AnthropicAI as a Member of Technical Staff. My defense slides ⬇️

@suvarna_ashima thank you! 🥳

Congratulations Dr. Bansal!! It has been an honor to stand beside you and watch you become everything you’ve worked for. I am so deeply proud of you, today and always.🌻

@adityagrover_ @kaiwei_chang Very grateful for your guidance and for learning so much from you!

Congrats, @hbXNov! Absolute delight to have co-advised Hritik with @kaiwei_chang. Back in 2022, Hritik had the foresight to use synthetic data to advance multimodal AI well before it became mainstream. His extensive body of PhD work since then have been instrumental in building systems that seamlessly reason across modalities.

Slides: docs.google.com/presentation/d…

Finally defended my Ph.D. thesis! 🥳 A very warm thank you to my family, friends, and advisors — @kaiwei_chang, @adityagrover_, @VioletNPeng, and Hongjing Lu. Next, I will be joining @AnthropicAI as a Member of Technical Staff. My defense slides ⬇️

RL optimizes weights. Evolution optimizes contexts. What if we combine RL and Evolutionary Algorithm (EA) into a new paradigm of LLM self-improvement? In "Evolutionary System Prompt Learning for Reinforcement Learning in LLMs", we show that RL and EA are deeply synergistic.

This paper is accepted to #CVPR2026! Link: arxiv.org/abs/2510.12225

Code: github.com/Hritikbansal/D…

Lastly, the blog also contains a personal account of how this project came into existence, and my impressions of changing research paradigms: "From Research 'On' AI to Research 'With' AI." Blog: huggingface.co/blog/hbXNov/de…

New blog 📢 Can we extract dense advantages without new annotations or models in GRPO? The answer is YES! 💡Answer correctness splits rollouts into positives and negatives. Just upweight positive tokens which differ significantly from the negative tokens! 🧵👇

@kaiwei_chang @baharanm Congratulations, great effort!

Today, we launched the UCLA DataX Center for AI Technology, co-directed by @baharanm and me. The center is dedicated to advancing the foundations of AI technology and enabling trustworthy real-world AI applications by bringing together researchers across disciplines. At the kickoff event, I had the pleasure of speaking about “Recent Advances in Multimodal Large Language Models and Their Applications,” discussing our recent multimodal LLM research that expands AI capabilities in mathematical reasoning, healthcare, and agentic AI. I also discussed current limitations of these models and potential paths forward. If you’re interested, please see the talk slides below. bit.ly/3ZIfWLD Thank you to everyone who joined us to celebrate this launch, and the panelists, Guy Van den Broeck @guyvdb , @YuchenCui1 @ElisaKreiss, and Karen McKinnon, for the fruitful discussion. We’re excited to grow this center into a hub for collaboration. datax.ucla.edu/news-events/ev…

@Yihe__Deng @WeiWang1973 @kaiwei_chang @baharanm @adityagrover_ congrats!

Finished my PhD defense this week! Immensely grateful to my advisor @WeiWang1973 and committee @kaiwei_chang @baharanm @adityagrover_ for their guidance and support over these years 🙏

New work💡: "EMA Policy Gradient: Taming Reinforcement Learning for LLMs with EMA Anchor and Top-k KL" Two ideas, both minimal, both effective: 🚀 use a target network (EMA) for reference policy 🚀 Top-k KL that works like knowledge distillation but remains unbiased at any k

@iclr_conf @clarkipeng @YonatanBitton @kaiwei_chang @adityagrover_ paper: arxiv.org/abs/2503.06800

VideoPhy-2 is accepted at @iclr_conf! great work by the talented undergrad @clarkipeng, my friend @YonatanBitton, Roman, and advisors @kaiwei_chang @adityagrover_ 😇