I am happy of the collab with the group of Elisa Ricci, w/ a paper led by Davide Berasi (coming from his brilliant MSc thesis) and co-authors Matteo Farina and Massimiliano Mancini! 📰"Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models"

🚨 CVPR main conference decisions are out, unlucky? Do not despair! The MULA workshop deadline is March 12, 2025. If you want to experience Nashville and present your work, don’t miss this opportunity! 🎸📅 #CVPR2025 More info here 👇🏻 mula-workshop.github.io

Definitely a bit late on this one, but 2/2 papers accepted at #CVPR2025 (one of which is a highlight ✨) I promised myself I'd post a thread about them the day I can get up as soon as my alarm rings... probably gonna take a while, stay tuned

🚀New Paper! arxiv.org/abs/2504.07086 Everyone’s celebrating rapid progress in math reasoning with RL/SFT. But how real is this progress? We re-evaluated recently released popular reasoning models—and found reported gains often vanish under rigorous testing!! 👀 🧵👇

Cool to be in #CVPR2025's list as well 🙏 Not gonna tell my mom this time tho...

What if we stopped treating image classification like a multiple-choice quiz… …and just asked the model: "What’s in this image?" Our paper on open-world classification with LMMs got into #ICCV2025! 🎉🌺 Let’s talk failures, insights, and flipping mistakes 👇

🚀New Paper arxiv.org/abs/2510.20860 We conduct a systematic data-centric study for speech-language pretraining, to improve end-to-end spoken-QA! 🎙️🤖 Using our data-centric insights, we pretrain a 3.8B SpeechLM (called SpeLangy) outperforming 3x larger models! 🧵👇

🚨Current data curation results in the creation of static datasets and the use of model-based filters that induce many biases. Can we fix this? We propose ✨CABS✨, a flexible concept-aware online batch curation method that improves CLIP pretraining! arxiv.org/abs/2511.20643 🧵👇

🚀New Paper arxiv.org/abs/2511.20643 Most data-curation for contrastive VLMs is offline and concept-agnostic! We introduce online, concept-aware batch sampling (CABS), enabling flexible and targeted curation, yielding ⬆️boosts over IID sampling (+7-9% on ImageNet + ret)! 🧵👇