Professor of Electrical and Computer Engineering at Tel Aviv University. On Sabbatical at Apple AI Research. Interested in AI, multimodal and signal processing
Iโm excited to be spending my sabbatical at #Apple, collaborating with Vladlen Koltun and the team on a range of cutting-edge AI projects. Looking forward to this new chapter!
@yoavgo When you supervise students, the accepted papers (for the good and the bad of it) are quite important for their future career either in academy or industry. So at least on my end, the excitement is for the students as I know how important it is for them
@ChrSzegedy I personally think that any arxiv submission should receive an AI review and the authors should decide if they still post their paper given the review or first revise and then submit with their answer to the review. Will fix a lot of slops
fresh out of the oven ๐ฅโจ
see the two videos below? with the 'Semantic Progress Function' (SPF) you can finally ANALYSE (and control!) the semantic progression of First-to-Last-Frame video generation.
Excited to share our work accepted to #SIGGRAPH2026 ! Video generation models struggle with something few talk about: their transformations don't evolve smoothly. You get long boring stretches... then a sudden semantic jump where everything "catches up" at once.
super excited this it out! lots of fun insights, tips, and tricks on how all parts of the RL stack, from policy gradient in general, to policy staleness, to clipping, to numerics, all modulate entropy + some principled control strategies to stabilize!
Our ICLR 2026 paper "Entropy-Preserving Reinforcement Learning" is now on arXiv.
We strike at one of the core RL issues: exploration in action space.
We study why token distribution entropy collapses in LLM post-training preventing further exploration and we propose fixes!
Introducing Look Where It Matters โ High-Resolution Crops Retrieval for Efficient VLMs.
VLMs don't need to process full high-res images. AwaRes uses tool-calling to retrieve only the high-res regions needed to answer a given query๐งต
https://t.co/JCiHCykcRP https://t.co/xqIYzmKeHP
๐๏ธ Introducing ID-LoRA: the first open-source model to jointly generate a video with a person's appearance and voice in a single pass from just a reference image + short audio clip. No more cascaded pipelines where the audio can't follow your prompt.
https://t.co/AZu9FfbzWG