Jayneel Parekh

ML/AI Researcher @Sorbonne_Univ_ | PhD in ML @telecomparis | EE Undergrad @iitbombay | XAI, Trustworthy AI, Generative Models, CV, Audio

[Jayneel Parekh @TrueBeerus]

🎉Happy to share our #NeurIPS2025 paper Learn-to-Steer (L2S). Offers flexible and better steering control for MLLMs than standard mean-steering approaches! Congrats @PKhayatan, @MustafaShukor1, Arnaud, Alasdair, and @quobbe!

- Arxiv: https://t.co/O3jMSwZXVO

[Thomas Fel @thomas_fel_]

Our work on Block-Sparse Featurizer is out 🧊 :)

We revive an old idea from the structured sparsity literature and use it to carve activation space into meaningful regions.

It's a first concrete answer to the question our concept manifolds work left open ! :)

[Sophia Sirko-Galouchenko @sophia_sirko]

1/n New paper - V-GIFT 🎁

Self-supervised tasks like rotation prediction or colorization were big in 2018.
Do they still matter?

Yes.
We turn them into visual instruction tuning data for MLLMs.

Result: models rely more on the image and perform better on vision tasks 👀

[Mustafa Shukor @MustafaShukor1]

We’re releasing Action100M: the largest open dataset of ~15 years of video with dense action + caption annotations.
It's a key ingredient behind VL-JEPA, now open to fuel the next generation of VLMs, World Models, and Robotics policies.

Dataset on HF: https://t.co/bUs67QyOdd

[Mustafa Shukor @MustafaShukor1]

VL-JEPA is out! A non-generative vision-language model, based on JEPA. Different from typical data-space autoregressive VLMs, VL-JEPA is trained to predict continuous embeddings in the latent space. (1/N)

[Pegah Khayatan @PKhayatan]

ELLIS Unconference at #EurIPS 🚀 Attended the Theory of Explainable ML Workshop—packed with ideas: B-Cosification, minimal+sufficient explanations, and concept-based insights. Great poster chats too!
Tomorrow we present our work on input-dependent MLLM steering — Poster #108

[Mustafa Shukor @MustafaShukor1]

We have a simple plan: a straight shot toward general-purpose robots. Whatever their form factor, humanoid or otherwise.

[Jayneel Parekh @TrueBeerus]

@LoickCh @PKhayatan @MustafaShukor1 @quobbe Thanks Loïck! Congrats for JAFAR and release of NAF too!

[Jayneel Parekh @TrueBeerus]

🎉Happy to share our #NeurIPS2025 paper Learn-to-Steer (L2S). Offers flexible and better steering control for MLLMs than standard mean-steering approaches! Congrats @PKhayatan, @MustafaShukor1, Arnaud, Alasdair, and @quobbe!

- Arxiv: https://t.co/O3jMSwZXVO

[Loïck Ch - 🇺🇦🇫🇷🇪🇺 @LoickCh]

Need pixel-level features from your backbone (DINOv3, CLIP, RADIO, FRANCA...)?
🚀Introducing NAF: A universal, zero-shot feature upsampler.
It turns low-res ViT features into pixel-perfect maps.

-⚡ Model-agnostic
-🥇 SoTA results
-🚀 4× faster than SoTA
-📈 Scales up to 2K res.

[Jayneel Parekh @TrueBeerus]

You can catch me and @PKhayatan during NeurIPS@Paris. @PKhayatan will also be at @EurIPSConf presenting the paper

Blog/Project page for further reading: https://t.co/WC5DGYoVg9

Code: https://t.co/TfpbsXj8eV

[Paul Couairon @PaulCouairon]

Remember JAFAR? We've just dropped its successor NAF!🎉

No need to retrain it, it is now zero-shot transferable to any backbone.

To put it simply, NAF is better, more efficient, and scalable than the things before!

[valeo.ai @valeoai]

Thrilled to have @TrueBeerus visiting us over a break in the CVPR sprint to talk about his exciting works in analyzing MLLMs representations. Check out his great work https://t.co/HlznK3vBRs

[Goodfire @GoodfireAI]

New research: are prompting and activation steering just two sides of the same coin?

@EricBigelow @danielwurgaft @EkdeepL and coauthors argue they are: ICL and steering have formally equivalent effects. (1/4)

[Jayneel Parekh @TrueBeerus]

Currently in Hawaii for #ICCV25
Will present our work "Analyzing Fine-tuning Representation Shift for Multimodal LLMs Steering" during Poster Session 1 tomorrow (Poster#200, Oct 21, 11:30AM-1:30PM HST). @PKhayatan, @MustafaShukor1, Arnaud Dapogny, @quobbe https://t.co/uvKJHfKD6J

[Sophia Sirko-Galouchenko @sophia_sirko]

Come say hi to our poster October 21st at 11:45 poster session 1 (#399)! We introduce unsupervised post-training of ViTs that enhances dense features for in-context tasks.
First conference as a PhD student, really excited to meet new people.

[Jayneel Parekh @TrueBeerus]

If you have any questions about state of explainability research, this can be a chance to get it answered by experts! Doesn't matter if it's something very simple or bizarre. Rather, hard-hitting questions are encouraged. You can also submit anonymously!

[Sweta Mahajan @CVPR2026 @SwetaMahajan1]

🚨 Call for Questions! 🚨

We are inviting the community and the stakeholders to submit questions, which will be discussed with our experts at the workshop! 🎤💡

👉 Submit your questions: https://t.co/ci9GyZWaZG

Workshop: https://t.co/dSofaHWmss

#ICCV2025 @ICCVConference #eXCV

[Sweta Mahajan @CVPR2026 @SwetaMahajan1]

🚨Deadline Approaching! 🚨⛰️🏖️

Non-Proceedings track closes in 2 days!

Be sure to submit on time!

We are awaiting your submissions!

More info at: https://t.co/lfmA0cMXFl

#ICCV2025 @ICCVConference #eXCV

[Sukrut Rao @sukrutrao]

Are you an XAI researcher attending #ICCV2025? Submit your recently published work — at CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, AAAI etc. — to the eXCV Workshop for the opportunity to further showcase your work! Published papers can be submitted as is, no rewriting necessary.

[Sweta Mahajan @CVPR2026 @SwetaMahajan1]

🚨Deadline Extension Alert!

Our Non-proceedings track is open till August 15th for the eXCV workshop at ICCV.

Our nectar track accepts published papers, as is.

More info at: https://t.co/lfmA0cMXFl

@ICCVConference #ICCV2025

[Robin Hesse @robinhesse_]

Got a strong XAI paper rejected from ICCV? Submit it to our ICCV eXCV Workshop today—we welcome high-quality work!
🗓️ Submissions open until June 26 AoE.
📄 Got accepted to ICCV? Congrats! Consider our non-proceedings track.
#ICCV2025 @ICCVConference