Adhiraj Ghosh ✈️ ACL 2025
@adhiraj_ghosh98
Followers
258
Following
371
Media
8
Statuses
57
ELLIS PhD @uni_tue | vision-language & data-centric ML @bethgelab 🦋: https://t.co/Q03vvJFIPw
Tübingen, Deutschland
Joined April 2024
🏆ONEBench accepted to ACL main! ✨.Stay tuned for the official leaderboard and real-time personalised benchmarking release!. If you’re attending ACL or are generally interested in the future of foundation model benchmarking, happy to talk!. #ACL2025NLP #ACL2025.@aclmeeting.
🚨Looking to test your foundation model on an arbitrary and open-ended set of capabilities, not explicitly captured by static benchmarks? . Check out ONEBench, part of my MS thesis, where we show how sample-level evaluation is the solution. 🔎. 🧵👇
1
1
10
RT @benno_krojer: This is precisely why we released MVPBench to measure genuine video understanding with minimal video pairs. Otherwise we….
arxiv.org
Existing benchmarks for assessing the spatio-temporal understanding and reasoning abilities of video language models are susceptible to score inflation due to the presence of shortcut solutions...
0
2
0
RT @YungSungChuang: Scaling CLIP on English-only data is outdated now…. 🌍We built CLIP data curation pipeline for 300+ languages.🇬🇧We train….
0
80
0
RT @sbdzdz: If you're in Vienna for ACL, @adhiraj_ghosh98 and I will be presenting our work on benchmarking language and vision-language mo….
bethgelab.github.io
ONEBench: a new paradigm for open-ended benchmarking and evaluation of foundation models, aggregating sample-level tests across datasets.
0
4
0
RT @KarelDoostrlnck: The fine folks at HuggingFace used our APO-zero algorithm to train a very efficient reasoning LM. Awesome 🙌 !.
0
6
0
RT @thao_nguyen26: Web data, the “fossil fuel of AI”, is being exhausted. What’s next?🤔.We propose Recycling the Web to break the data wall….
0
62
0
RT @AmyPrb: 🚨 New paper!. Exciting progress in GRPO variants, smarter training strategies, and curated datasets showing impressive improvem….
0
4
0
RT @debsarkar_sayan: 🏆 CrossOver is accepted as a 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁 at @CVPR #CVPR2025! ✨.💻 Fully open-sourced code with all pre-trained checkpoint….
github.com
[CVPR 2025, Highlight] CrossOver: 3D Scene Cross-Modal Alignment - GradientSpaces/CrossOver
0
4
0
RT @fededagos: 🚨 New paper alert! 🚨.We’ve just launched openretina, an open-source framework for collaborative retina modeling across datas….
0
9
0
RT @AmyPrb: LMs excel at solving problems (~48% success) but falter at debunking them (<9% counterexample rate)! . Could form an AI Brandol….
0
1
0
If you needed a reason to visit Nashville in June🔥.
🎉 Excited to share our latest work, CrossOver: 3D Scene Cross-Modal Alignment, accepted to #CVPR2025 🌐✨. We learn a unified, modality-agnostic embedding space, enabling seamless scene-level alignment across multiple modalities — no semantic annotations needed!🚀
0
0
1
Valuable multimodal training data contribution and very thorough experimentation🎉. Interesting that fine tuning the visual encoder doesn’t help in such tasks, seems to be aligned with common wisdom in the LMM world.
[1/11] Many recent studies have shown that current multimodal LLMs (MLLMs) struggle with low-level visual perception (LLVP) — the ability to precisely describe the fine-grained/geometric details of an image. How can we do better?. Introducing Euclid, our first study at improving
0
0
3
RT @AmyPrb: 🚨Looking for open problems machine unlearning for AI safety? . We provide a deep dive into the nuances of removing harmful know….
0
2
0
RT @JieyuZhang20: Excited to share my intern project at Salesforce Research! Huge thanks to everyone on the team!!.
0
15
0
Used LMMs-Eval to build ONEBench(: fantastic team who have diligently helped me with queries and access to data over the past 6 months. Glad to see it gain so much traction!.
arxiv.org
Traditional fixed test sets fall short in evaluating open-ended capabilities of foundation models. To address this, we propose ONEBench(OpeN-Ended Benchmarking), a new testing paradigm that...
🚀LMMs-Eval🚀 has reached 2.2K stars with 60+ contributors from the community!. - Repo: - Dataset @huggingface : * Join us to build a standardized evaluation toolkit for large multimodal models (image, video and audio)🤩.
0
2
9
RT @BoLi68567011: Also recommend the open-source code for SAE in multimodal models.
github.com
[ICCV 2025] Auto Interpretation Pipeline and many other functionalities for Multimodal SAE Analysis. - EvolvingLMMs-Lab/multimodal-sae
0
16
0
No surprises whatsoever but congratulations!.
🫡 accepted to TACL.
0
0
1