Aldozzzzz's Avatar

Aldozzzzz

@az-mtl

Applied ML Research @mila-quebec.bsky.social, currently focused on VLMs and climate. Alumnus of @polymtl.bsky.social. Travel, Photography & Amateur Music Composer. https://alzaia.github.io/ He/him. πŸ‡«πŸ‡·πŸ‡¨πŸ‡¦

545
Followers
735
Following
328
Posts
07.11.2024
Joined
Posts Following

Latest posts by Aldozzzzz @az-mtl

A boring Mario Party game. No Zelda, no 3D Mario…

08.03.2026 15:02 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0
Preview
CLIP Is Shortsighted: Paying Attention Beyond the First Sentence CLIP models learn transferable multi-modal features via image-text contrastive learning on internet-scale data. They are widely used in zero-shot classification, multi-modal retrieval, text-to-image d...

Our freshly accepted CVPR 2026 paper is up on arxiv 😍. Project page, open code and more detailed post coming soon!

« CLIP is Shortsighted: Paying Attention Beyond the First Sentence ».

#CVPR

27.02.2026 03:40 πŸ‘ 5 πŸ” 2 πŸ’¬ 0 πŸ“Œ 1
Video thumbnail

Positive vibes only. #pokopia

08.03.2026 01:52 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

πŸ˜‚πŸ˜‚πŸ˜‚

08.03.2026 00:46 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Video thumbnail

Thank you Nintendo for always delivering at the right time. Animal Crossing for the pandemic, Pokopia for WW3 😭.

06.03.2026 22:58 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 1

+ no AI generated references πŸ˜›

05.03.2026 23:40 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

This is very cool. I kinda wish one of the big conferences would make a non-LLM submission policy for the main track just for one year. Would be fun πŸ˜›.

05.03.2026 23:38 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

I just hide and cry about it 😁.

05.03.2026 21:40 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Project page now available! DeBias-CLIP is now the new SOTA for image-to-long-caption retrieval!

trailab.github.io/DeBias-CLIP-...

03.03.2026 23:55 πŸ‘ 4 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Post image Post image

πŸ˜‚πŸ˜‚πŸ˜‚πŸ˜­πŸ˜­πŸ˜­

28.02.2026 16:27 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

😍😍😍

28.02.2026 15:45 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Preview
CLIP Is Shortsighted: Paying Attention Beyond the First Sentence CLIP models learn transferable multi-modal features via image-text contrastive learning on internet-scale data. They are widely used in zero-shot classification, multi-modal retrieval, text-to-image d...

Our freshly accepted CVPR 2026 paper is up on arxiv 😍. Project page, open code and more detailed post coming soon!

« CLIP is Shortsighted: Paying Attention Beyond the First Sentence ».

#CVPR

27.02.2026 03:40 πŸ‘ 5 πŸ” 2 πŸ’¬ 0 πŸ“Œ 1
Post image

In this amazing multidisciplinary collaboration, we report our early experience with the @openclaw-x.bsky.social ->

23.02.2026 23:32 πŸ‘ 40 πŸ” 21 πŸ’¬ 1 πŸ“Œ 9
Preview
Memorization vs. generalization in deep learning: implicit biases, benign overfitting, and more Or: how I learned to stop worrying and love the memorization

What is the relationship between memorization and generalization in AI? Is there a fundamental tradeoff? In infinitefaculty.substack.com/p/memorizati... I’ve reviewed some of the evolving perspectives on memorization & generalization in machine learning, from classic perspectives through LLMs.

18.02.2026 15:54 πŸ‘ 134 πŸ” 27 πŸ’¬ 4 πŸ“Œ 5
Post image
23.02.2026 23:09 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Post image

Inspiring and on point quote for 2026 πŸ₯².

22.02.2026 18:34 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

Haven’t finished it yet. Played last winter on NSO πŸ˜….

22.02.2026 03:18 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

BOTW ~ TOTK ~ Skyward Sword πŸ’šπŸ’šπŸ’š

22.02.2026 03:07 πŸ‘ 0 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

He’s so freaking cringe 😬. Is this real life? 🀯🫠

21.02.2026 22:37 πŸ‘ 2 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

Honestly a pretty simple solution: we force these models to pay attention to the whole text sequence πŸ‘€. The motivation of this work was that all previous models were mostly focusing on the first few tokens (the « summaryΒ Β» sentence). Details are in later parts of the caption typically πŸ‘€.

21.02.2026 19:10 πŸ‘ 1 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

This was legit the first time I was feeling confident about a big AI conf. paper! Initial reviews were already 3/3 accepts, and we managed to get final scores of 5/5/6!

Teaser: we’re now the new SOTA (by a huge margin) for image-to-long-text CLIP-style retrieval πŸ‘€.

21.02.2026 17:51 πŸ‘ 5 πŸ” 0 πŸ’¬ 1 πŸ“Œ 0

Accepted at CVPR! More info to come soon (we need to work on our project page and all πŸ˜…). πŸ‘€

21.02.2026 08:33 πŸ‘ 5 πŸ” 0 πŸ’¬ 0 πŸ“Œ 1
Preview
a cartoon of a man sitting in front of a computer screen that says vertek ALT: a cartoon of a man sitting in front of a computer screen that says vertek

Me refreshing OpenReview every 10 seconds for CVPR decisions πŸ‘€.

20.02.2026 18:51 πŸ‘ 2 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Preview
a cartoon drawing of a llama wearing a purple shirt with a blue and green logo on it ALT: a cartoon drawing of a llama wearing a purple shirt with a blue and green logo on it
17.02.2026 00:32 πŸ‘ 2 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

😩😩😩

16.02.2026 02:28 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0

5 days to go until final CVPR decisions πŸ‘€πŸ‘€πŸ‘€. #cvpr

15.02.2026 18:47 πŸ‘ 0 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
"Moderna's CEO announced the company will no longer invest in new Phase 3 vaccine trials for infectious diseases: 'You cannot make a return on investment if you don't have access to the U.S. market. Vaccines for Epstein-Barr virus, herpes, and shingles have been shelved.”
9:12 AM
Feb 12, 2026
604.8K Views

"Moderna's CEO announced the company will no longer invest in new Phase 3 vaccine trials for infectious diseases: 'You cannot make a return on investment if you don't have access to the U.S. market. Vaccines for Epstein-Barr virus, herpes, and shingles have been shelved.” 9:12 AM Feb 12, 2026 604.8K Views

In terms of what cures are being lost:

- Epstein-Barr virus is perhaps the major trigger for multiple sclerosis
- herpes simplex virus causes cold sores, genital herpes, infections in babies, deadly meningitis
- shingles virus causes an intensely painful disease

13.02.2026 05:44 πŸ‘ 12539 πŸ” 6207 πŸ’¬ 443 πŸ“Œ 783

You still have time to apply to the CVPR (horror) art submission πŸ˜‚.

08.02.2026 23:28 πŸ‘ 2 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0
Post image

It took me weeks, but finally it's there: an overlong blogpost on synthetic pretraining. vintagedata.org/blog/posts/s...

01.02.2026 17:53 πŸ‘ 86 πŸ” 21 πŸ’¬ 3 πŸ“Œ 2

Very very cool and important blog!

01.02.2026 23:46 πŸ‘ 1 πŸ” 0 πŸ’¬ 0 πŸ“Œ 0