PostDoc @VIB-KULeuven | Prev: PhD@KUL (ELLIS), UvA, @iiit_hyderabad, @iiscbangalore
Leuven
Joined June 2010
- Tweets286
- Following251
- Followers98
- Likes450
Abhishek Jha retweeted
This is a result I’ve dreamt about for many years: *unpaired translation between images and text*
I thought it might be only slightly possible, the kind of thing you have to really squint at. But Dominik proved this wrong. You don’t have to squint. Worth looking for yourself:
DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair.
It even works when the images and the captions come from different datasets.
Project page: dominik-schnaus.github.io/un…
⬇️
Abhishek Jha retweeted
#CoLLAs2026 was a very productive conference for our lab!
@ti_hess @JhaAbsk presented a poster on co-observation, a third facet of #continual_learning
@Atenrev presented:
- Putting a Face to Forgetting: cl meets mech interp
- Lifelong Representations: CSSL survey
Congrats! ❤️
Catastrophic forgetting and loss of plasticity are treated as the two obstacles in continual learning.
We point at a third factor: data co-observation.
@ti_hess presents our poster tomorrow at #CoLLAs2026.
arxiv.org/abs/2608.18803
Abhishek Jha retweeted
Catastrophic forgetting: your model "forgets" a task, and probing confirms the info is gone. Case closed... right?
Not so fast
We found that sometimes tests like probing don't tell the full story. Features are still there, just misaligned or too crowded.
Let's dive in!🧵
Abhishek Jha retweeted
Our colleague Baptiste has presented “Multivariate Distributional Reinforcement Learning Using Sliced Divergences” at ICML 2026!
Sliced divergences make multivariate distributional RL tractable, but only some of them are suitable for real one-sample TD.
🧵Paper link below👇
Abhishek Jha retweeted
Want to improve content safety and NSFW detection in CLIP? Hyperbolic geometry can make this possible. Check out our #CVPR2025 paper, Hyperbolic Safety-Aware Vision-Language Models.
With: @TobiaP93332, @PascalMettes, @lorenzo_baraldi, @ricucch #ELLISforEurope #Ellis_Amsterdam
Abhishek Jha retweeted
Today at #ICCV2023 me and @dusangrujicic will present our latest work - Multimodal Distillation for Egocentric Action Recognition. You can find us from 14:30 to 16:30 (Poster 087). Feel free to reach out! 😀
Abhishek Jha retweeted
Contributed talk on “Maximum Class Separation as Inductive Bias in One Matrix” by Tejaswi Kasarla #WiMLworkshop #NeurIPS2022
@NeurIPSConf
Abhishek Jha retweeted
Our #NeurIPS2022 paper is selected as Oral! I'm beyond delighted! 🎉
You can visit our poster at @WiMLworkshop and the conference in-person; and at the deep-dive panel during virtual week!
@gjburghouts @MvanSpengler @ElisevanderPol @ricucch @PascalMettes
Looking to boost classification, long-tailed recognition, OOD/open-set recognition with single line of code?
Read our new work, Maximum Class Separation as Inductive Bias in One Matrix: arxiv.org/abs/2206.08704
w/ @gjburghouts,@MvanSpengler,@ElisevanderPol,@ricucch,@PascalMettes
ALT single line of code to improve classification, long-tailed recognition, out-of-distribution detection and open-set recognition.
Abhishek Jha retweeted
We organized a joint workshop with Tinne Tuytelaars' lab at KU Leuven. We exchanged ideas on continual learning, object-centric representations, open set recognition and benchmarks. Well-spent two days in Cologne!
Abhishek Jha retweeted
Finally, my fav result: it's possible to train models on disjoint and biased datasets, then merge them together in weight space.
Eg, you have some data in US, some in EU. Can't mix data due to GDPR etc. Train separate models, merge weights -> generalize to the combined dataset!
Abhishek Jha retweeted
Adversarial examples, originally observed ~2014, were fascinating cases showing deep nets can be super sensitive to small input perturbations. Since then a lot has happened! Did we solve this failure mode of AI models or have we hit a wall? A thread 🧵
Abhishek Jha retweeted
Looking to boost classification, long-tailed recognition, OOD/open-set recognition with single line of code?
Read our new work, Maximum Class Separation as Inductive Bias in One Matrix: arxiv.org/abs/2206.08704
w/ @gjburghouts,@MvanSpengler,@ElisevanderPol,@ricucch,@PascalMettes
ALT single line of code to improve classification, long-tailed recognition, out-of-distribution detection and open-set recognition.
Abhishek Jha retweeted
At #CVPR2022? I have fully funded MSc and PhD positions available for fall 2023.
Interested? Reach out via DM and let’s chat at the conference. – at New Orleans, LA
I wonder if the reviewer stats are similar to that of attendees: cvpr2022.thecvf.com/exhibito… @CVPR #CVPR2022
Abhishek Jha retweeted
I am looking to hire a good post doc at Center for Research in Computer Vision (CRCV) at University of Central Florida. If interested please send me your resume at shah@crcv.ucf.edu
Abhishek Jha retweeted
Super excited to share this new joint work with @jm_alexia and @chrisjpal !
I'm super excited to share this new joint work with @VikramVoleti and @chrisjpal where we show that a single diffusion model can solve many video tasks through a masking scheme! 🧙♂️
Blog: ajolicoeur.wordpress.com/?p=…
Website: mask-cond-video-diffusion.gi…
Arxiv: arxiv.org/abs/2205.09853
Abhishek Jha retweeted
I probably should have written this years ago, but here are some MLOps principles I think every ML platform (codebase, data management platform) should have: 1/n
