Kavana Venkatesh

🚀 Introducing PhysicsAgentABM! We propose a novel Physics-Guided Generative Agent Based Modeling framework for scalable and calibrated simulation of complex social systems! ⚡ 6.7x fewer API calls, 2.1-3.4x cost reduction 🎯 Superior temporal accuracy across all domains (EETE, ET-F1) 📊 Calibrated uncertainty (near-diagonal reliability curves) 🔄 Robust to regime shifts (policy shocks, market crashes, viral events) Huge kudos to my collaborator @inhan_y , Prof. @LiJundong and my advisor @jiamingcui1997!

🔗 Checkout our preprint below 👇 📄 Paper: arxiv.org/abs/2602.06030 We welcome your comments and feedback!

💹Cost & Scalability Flat GABM vs PhysicsAgentABM (N=1000): API calls: 8,250 → 1,233 (6.7× reduction) Tokens: 2.3M → 0.49–0.79M Cost: 2.1–3.4× lower Latency: 7.5–12.5× faster ET-F1: improves from 0.62 → 0.81 Reasoning cost scales with clusters, not agents.

🚀 Introducing PhysicsAgentABM! We propose a novel Physics-Guided Generative Agent Based Modeling framework for scalable and calibrated simulation of complex social systems! ⚡ 6.7x fewer API calls, 2.1-3.4x cost reduction 🎯 Superior temporal accuracy across all domains (EETE, ET-F1) 📊 Calibrated uncertainty (near-diagonal reliability curves) 🔄 Robust to regime shifts (policy shocks, market crashes, viral events) Huge kudos to my collaborator @inhan_y , Prof. @LiJundong and my advisor @jiamingcui1997!

@RevanthAtmakuri Thanks for sharing our work!

PhysicsAgentABM blends LLMs and ABMs, clustering agents for smarter, cheaper simulations 💻🐈👩. It nails event timing and calibration, cutting LLM costs 6-8x 👏. Great for complex systems, but watch for oversimplification in nuanced behaviours. - AI arxiv.org/abs/2602.06030

So grateful for the incredible students and alumni of GEMLAB! Together, we published 4 main conference papers + 4 workshop papers at @NeurIPSConf. You’re all absolute gems 💎

My amazing students Connor, Matt and Kavana are presenting our paper at poster#5319 right now! @NeurIPSConf

@d_yesiltepe @docmilanfar @yusuf_dalva @tunahansalih 💪

Today we met the legendary @docmilanfar and in the back @yusuf_dalva looked terrified by his sudden appearance :)

Kavana @kavanav2912 and Connor presenting our paper at poster #5402 right now! @NeurIPSConf

GEMLAB is heading to @NeurIPSConf! 🎉 We’re presenting 4 main conference papers (including 1 spotlight ⭐) + 4 workshop papers. Come say hi & check out our work! Details below 👇

Huge kudos to my amazing students for their hard work and dedication: Hidir @d_yesiltepe, Yusuf @yusuf_dalva , Connor, Tuna @tunahansalih, Matthew, Kavana @kavanav2912, Kiymet @akdemir_kiymet, and Tahira @TKazimi415! 🎉 We'll all be at @NeurIPSConf, come say hi!

Heading to NeurIPS to present two papers! 🖼️ CREA: A Collaborative Multi‑Agent Framework for Creative Image Editing and Generation 🌐 Project Page: crea-diffusion.github.io 👩‍🔧Personalized Image Editing in Text‑to‑Image Diffusion Models via Collaborative Direct Preference Optimization 🌐 Project Page: personalized-editing.github.io If you want to chat about Multi-Agent Safety, Multimodal GenAI and applications, let's get a ☕☺️ I am also looking for Research Internships for Summer 2026 and would appreciate any leads! See you 🔜 in San Diego! 😃 #NeurIPS2025 #NeurIPS #GenAI

🎊 Thrilled to share that 2 of my papers got into NeurIPS! Huge thanks to my collaborators and our amazing advisor, Dr. Pinar Yanardag @PINguAR at @SanghaniCtrVT! 🚀Can’t wait to present our work in San Diego this December! #NeurIPS2025

@yusuf_dalva @d_yesiltepe @PINguAR Amazing!! Many congratulations @yusuf_dalva!

Excited to share that LoRAShop has been accepted at #NeurIPS25 as a Spotlight! 🎉 Huge thanks to my collaborator @d_yesiltepe and my advisor @PINguAR 🙏 Code will be released soon, looking forward to connecting with the community in San Diego!