harini s i

Research @ Adobe | BITS Goa

[harini s i @hariniskumar10]

Paper: https://t.co/J6mFVUqEwX
Project page: https://t.co/qQ54JzumCM

See you at NeurIPS 2026!

[harini s i @hariniskumar10]

Excited to share that Experigen has been accepted to NeurIPS as a Spotlight! 🎉

When we introduced Experigen, our goal was to make scientific discovery from observational data faster: turning messy real-world data into hypotheses that can be statistically tested and validated.

[harini s i @hariniskumar10]

Understanding human behavior is one of the oldest problems in science. The web has given us more data to study this than ever before: social media posts, reviews, debates, images. Yet science still lacks a framework to convert this data into meaningful hypotheses. 🧵

[harini s i @hariniskumar10]

Huge thanks to my amazing co-authors: Jishu, @someshsingh22, @YamanKSingla, @smta56, @ddoermann, @RatnRajiv, @KbalajiTweets , Jitendra.

If you’re interested in AI for social science, human behavior modeling, or AI-driven scientific discovery, I’d love to chat!

[harini s i @hariniskumar10]

I’ll be presenting our paper on image personalization using personas at Poster Session 4, Booth #264! 🎉

If you’re attending, come say hi! And if you’re interested in personas, LLMs, or modeling human behavior, I’d love to chat.

📩 [email protected]

#alignment #persona #LLM

[Somesh @someshsingh22]

We're also releasing ZIP-Bench: 1,500 users, 40K images.

Huge thanks to my co-authors @hariniskumar10 and mentors @YamanKSingla, @ddoermann & @RatnRajiv

If you are📍 At #ECCV2026 in Malmö Come to our poster — [session 4 board #264]. If you work on personas, generative models, or human behavior, come say hi 👋

🔗 https://t.co/ekGCFHN29k

[Somesh @someshsingh22]

The scientific cycle of observing, hypothesizing, and experimenting underpins every discovery across data driven sciences including social science, marketing, healthcare, pathology, and human behavior.

Over the past decades, datasets and AI reasoning/coding capabilities have scaled dramatically. Yet the methods for forming and validating our theories have not kept up.

In this preprint, we introduce ExperiGen: a framework that automatically generates intervention-ready hypotheses from raw observational data, validated as novel, impactful, and worthy of investigation by expert social scientists, accelerating science from what took months to hours in under $10.

In collaboration with a Fortune 500 company, we conducted the first A/B test of an LLM-generated hypothesis, achieving a large effect size (OR = 3.44) with strong significance (p = 1e-6).
https://t.co/5hPtT37S9f

[harini s i @hariniskumar10]

To know more, read our paper: https://t.co/J6mFVUqEwX
Huge thanks to my amazing co-authors: Jishu SenGupta, @someshsingh22, @YamanKSingla, @smta56, @ddoermann, @RatnRajiv, @KbalajiTweets

[harini s i @hariniskumar10]

Understanding human behavior is one of the oldest problems in science. The web has given us more data to study this than ever before: social media posts, reviews, debates, images. Yet science still lacks a framework to convert this data into meaningful hypotheses. 🧵

[harini s i @hariniskumar10]

We also ran the first A/B test of LLM-generated hypotheses , with statistically significant results (p < 1e-6) & a +344% effect size.
Experigen automates the most time-consuming parts of scientific discovery — so researchers can focus on what matters: asking better questions.

[Somesh @someshsingh22]

LLMs will solve this in the coming week 😉
Never if you show an incomplete puzzle, either send the PGN or highlight the last move tho 🙃

[Sun, Fan-Yun ✈️ CoRL @sunfanyun]

Check us out at NeurIPS tomorrow! Unfortunately I can’t be there but @locross and Jonathan will present at
East Exhibit Hall A-C

[Somesh @someshsingh22]

It's really interesting to see the "human difficulty" of sharp positions combined with time pressure still evade most known statistical methods, Na1 was a genius move in hindsight. GG @DGukesh. Looks like SOTA engines still have a great scope ;) #DingGukesh

[harini s i @hariniskumar10]

My first NeurIPS paper! Do check it out 🥳

[Sun, Fan-Yun ✈️ CoRL @sunfanyun]

Training RL/robot policies requires extensive experience in the target environment, which is often difficult to obtain. How can we “distill” embodied policies from foundational models?

Introducing FactorSim! #NeurIPS2024
We show that by generating prompt-aligned simulations and training a policy on them without collecting any experience in the target environment, we can achieve zero-shot performance close to policies trained on millions of target environment experiences in many classic RL environments.

You can generate RL simulations on our project website: https://t.co/bQ95SkFjrS

More in 🧵

1/7

[SAiDL @SforAiDL]

Check out the paper published by SAiDL member @hariniskumar10, @gmshroff, Prof.Ashwin Srinivasan, Prayushi Faldum and Lovekesh Vig at the AAAI-2023 Workshop on Multimodal AI for Financial Forecasting.

https://t.co/HyyIMfIY32

[SAiDL @SforAiDL]

We are excited to present the speakers for "AI Symposium 2022", Sharut Gupta (@sharut_gupta), Ph.D. Candidate at @MIT_CSAIL and Rishab Khincha (@rishabkhincha), MSCS at @UTAustin. They will discuss MS - Ph.D. programs, including the application procedure.

[SAiDL @SforAiDL]

Continuing the trend, we have 4 selections from SAiDL for Google Summer of Code 2022. We would like to congratulate @twm_as, @hariniskumar10, @ShreyPandit2001, and @NeelayShah8.

[1/5]

[SAiDL @SforAiDL]

Check out this recent work by SAiDL member @someshsingh22 and @SwapnilDreams10, @YamanKSingla, @KbalajiTweets, @cchangyou, @RatnRajiv on generating data-free universal adversarial triggers for NLP systems published in AAAI-22.

https://t.co/rC0ZwYsyoX

1/3