Presenting at @icmlconf 🇰🇷! 🕑 Tue, Jul 7, 2–3:45 PM KST 🖼️ “Jailbreaking VLMs Through the Visual Modality” Frontier VLMs can be jailbroken by making them recover unsafe intent from visual context! At #ICML? Let’s talk about VLM safety. w/ @AharonAzulay, Zhuoyun Li, Atharv Mittal & @YGandelsman

Join us on July 28 for the Paris AI, ML and Computer Vision Meetup! Pre-registration is mandatory as seats are limited - hubs.ly/Q04nvGgn0 Talks will include: * Finetuning VLMs for domain specific tasks - Amine Belhakimi at GoPro * Computer Vision at Nanoscale - Detecting, Segmenting and Analyzing Nanoparticles in microscopic images - Atif Anwer at (Ex) University Bourgogne Europe * Towards a Resolution- and Modality-Agnostic Transformers for Earth Observation - Guillaume Astruc at Imagine - ENPC * Building Real-World Computer Vision Systems with Voxel51 - Harpreet Sahota at Voxel51 * Efficient Image Generation through Smarter Data, Objectives, and Alignment - Lucas Degeorge at Ecole Polytechnique - Ecole des Ponts - AMIAD *********** Level up your computer vision workflows with a free hands-on workshop for your team! Book a workshop: hubs.ly/Q04nvLHd0 These hands-on workshops are delivered by Voxel51 computer vision experts. Both virtual and in-person formats. * 60 min virtual workshop * Half-day onsite workshop * Full-day onsite workshop and hackathon #mcp #skills #computervision #ai #artificialintelligence #machinevision #machinelearning #physicalai

Looking for qualified NLP/ML folks who can do an emergency ARR/EMNLP review in the next 24-36h -- areas: multimodal/VLMs, RAG, hallucinations, toxicity/safety, humour, dialogue, remote sensing/vision-language, table reasoning; please DM me!

Modern VLMs arent just image captioners; they embed visual features into the LLM's context, letting the same model answer questions, read documents, and reason over charts in a single prompt. analyticsvidhya.com/blog/202…

Surprisingly, VLMs have a really hard time solving these tasks. Even very capable VLMs often ignore the visual context and instead latch onto the most salient object in the query: The same query paired with a different context should lead to a different interpretation, but often does not.

Disentangling Pictorial Cue Understanding from Language Bias in VLMs via Depth Ordering Task Yiqian Liu, Iuliia Kotseruba, John K. Tsotsos arxiv.org/abs/2607.01503 [𝚌𝚜.𝙲𝚅] 💬Accepted to ECCV 2026

I'm at #ICML2026! Reach out if you’d like to discuss hybrid models, efficient LLMs/VLMs, memory models, or related topics.

Ahh if only SpatialTransformerNet (not a transformer lol) actually worked at scale. They are the perfect match for this in theory, and beautiful. In practice instead we have VLMs going "hmm thinking hmmm let me use my zoom and crop tool here hmmm"

Want to understand how AI bridges the gap between text and vision? 👁️ ✍️ This deep dive breaks down Modern Vision Language Models (VLMs), how they work, and why they are transforming multimodal AI. Read the full breakdown on @AnalyticsVidhya 👇 analyticsvidhya.com/blog/202… #AI #MachineLearning #VLMs #ComputerVision #GenAI

On Wednesday morning I'll be presenting this poster about chain-of-thought consistency in VLMs finetuned with RLVR. This was done last summer at an internship with Apple, and will also have a presentation at the Apple Booth in the afternoon! (More details here: machinelearning.apple.com/up…) x.com/rosieyzh/status/203006…

Working on VLMs at the edge is what I do, so I quantized Qwen3.5-4B to NVFP4 for my own deployments. Figured it could be useful for others too, so I’m publishing it on Hugging Face. huggingface.co/sanskar003/Qw… #qwen #vlm