Dhruv Agarwal

✨✨ 2 papers accepted to #ICLR2025!! Hoping our work can move us a little closer to autonomous scientific discovery!🧬🥼 DiscoveryBench (openreview.net/forum?id=vyflg…) Searching for Optimal Solutions with LLMs via Bayesian Optimization (openreview.net/forum?id=aVfDr…)

Lots of great work on scaling test time compute yesterday @iclr_conf! We'll be at Hall 3 + Hall 2B #222 today at 10a-12:30p SGT talking about our take - BOPRO! Come say hi! 👋

Excited to share what I have been focusing on this year! Inference-time search to optimize Bayesian surprise pushes us towards long-horizon discovery! Introducing "AutoDS": Autonomous Discovery via Surprisal. "It can not only find the diamond in the rough, but also can rule out…

Great science starts with great questions. 🤔✨ Meet AutoDS—an AI that doesn’t just hunt for answers, it decides which questions are worth asking. 🧵

Hello @naaclmeeting attendees, 📢 This Friday, May 2, starting at 1 PM MST, I am hosting a live demo of our upcoming research tool: DataVoyager from @allen_ai. It’s designed for open-ended data analysis and discovery—you can chat with your uploaded datasets to generate…

if I was not already a fan of @allen_ai this structured approach of discovery with LLMs does the job.. had an awesome deep dive on the paper and approach with @dhruvagarwal17 at @iclr_conf

Sad to miss @iclr_conf this year, but @dhruvagarwal17 is going to post the flag for @allen_ai presenting DiscoveryBench, a benchmark to assess automated data-driven discovery! Catch up with him to know more about the work we are doing in the automated discovery space! I will be…

This will now appear in @iclr_conf (while we iron out some ethics reviews regarding sharing the datasets)!! #ICLR2025 Fantastic work, team! @allen_ai Updated version + some new releases coming soon! 🚀

Friends and colleagues! I'm on the academic job market in Fall 24/Spring 25. My research intersects NLP, IR, and HCI and I develop interactive and personalized models to aid workflows where knowledge workers find, consume, and produce information. Please help spread the word! 🧵

Can LLMs help accelerate the discovery of data-driven scientific hypotheses? 🧬📊 We benchmark this in DiscoveryBench: 264 discovery tasks from 6 scientific domains, from humanities to biology: arxiv.org/pdf/2407.01725… @allen_ai @ai2_aristo @surana_h @UMassAmherst

Excited to be in Mexico City 🇲🇽 for #NAACL2024 to present my work with @AmazonScience on using LLMs for knowledge graph question answering! Come say hi at our poster in Session 7 on June 19 at 11am! 🙂 Paper: aclanthology.org/2024.findings-… Repo: github.com/amazon-science…

You might have seen this already but if not, we are super excited with this new direction on hypothesis verification and discovery, grounded in data and code for reproducibility! Catch us in upcoming conferences (NAACL, ICML, ACL) to know more!! 🙌🏼

There's a mountain of great research and datasets that exists today — can AI help sort through the pile? We're excited to present DataVoyager, work by @mbodhisattwa and co-authors that aims to assist scientists in discovering new insights and verify known hypotheses…

✨🧬How can the process of scientific discovery be systematically automated to accelerate the expansion of the frontier of knowledge? We take a first step at answering this question in our new position paper. Very excited about this direction with some fantastic collaborators!!

We had the pleasure of hosting a fantastic lineup of speakers to talk about their research at the Machine Learning & Friends Lunch (MLFL) last Fall at @UMassAmherst! The talks are now available at youtube.com/channel/UC9Gp6…. Stay tuned for an equally exciting Spring 2024!

🚨 New Paper Alert 🚨 Students in knowledge distillation (KD) can outperform their teacher---we exploit this phenomena and train two students in parallel on disjoint data, then perform successive rounds of KD to incrementally improve the student(s).