Ali Vosoughi

🤖 Multimodal AI Research Scientist @Apple 🏢 Former @MSFTResearch @Bosch_AI | PhD @UofR 🎯 Audio-Visual Generation • Causal AI

[Ali Vosoughi @ali_vosough]

OSCaR is now public: project page, code, dataset, and checkpoints.

NAACL work on object state captioning and state-change representation for grounded video understanding.

Project: https://t.co/eX2UmHvBzV
Paper: https://t.co/HHOeY6T0JI
Dataset: https://t.co/th71UD9CHe

[Ali Vosoughi @ali_vosough]

Paper: https://t.co/pc7KCBSzEy
Dataset: https://t.co/qQu1Pv7CPo

To be presented at COLM in San Francisco this October.

[Ali Vosoughi @ali_vosough]

Our benchmark VERIFY was accepted at COLM 2026.

600 puzzles, each with a human-annotated reasoning path: we grade how models reason, not just what they answer.

Try this one. In our evaluation, most frontier models missed it.

[Kevin Lin @KevinQHLin]

https://t.co/trPSHlwFLT

[Yongyi Zang @yongyi_zang]

Something we did over the summer. One thing I’m particularly proud about this work is how easy the task could’ve been (literally just caption the images!), and how many extra steps we took just to make sure creators can actually use this system in their work, rather than just making it as a research POC. You’ll see in the paper how many ablations and extra design we did, as well as some choices we deliberately made despite objective metrics telling us otherwise.

The additional steps weren’t required to make it a valid paper, but are crucial to making it something people actually will be willing to use. Thanks so much for @ali_vosough and the team for working together on this!

[Ali Vosoughi @ali_vosough]

🚀 HUGE thanks for the mentorship of @yongyi_zang and Randal Leistikow for the opportunity, and great teamwork @rocksoul183 and Nathan Paek and my advisor @ChenliangXu ✨

[Ali Vosoughi @ali_vosough]

Excited to share our latest work: PromptReverb, a text-to-reverb generation system using latent flow matching! 🎧🎵
Paper: https://t.co/FziNqXl5hF

[Ali Vosoughi @ali_vosough]

This work was done at @smulelabs @smule, and we're preparing to open-source the model code and weights soon. More details in the paper!

[Ali Vosoughi @ali_vosough]

@PooyanFazli Congras🎉!

[Ali Vosoughi @ali_vosough]

🚀 Released AVE-2: A dataset for unified audio-video understanding + generation

Designed for training models that both perceive and synthesize across modalities.

Perfect for testing your next multimodal architecture.

🔗 https://t.co/DXEIYVLp38

#MultimodalAI #AudioVisual

[Ali Vosoughi @ali_vosough]

@ysbhalgat @CVPR Great to see activities for CVPR @ysbhalgat ! I'm also presenting our vision agents work and looking to connect with teams working on multimodal AI, especially audio-visual intelligence. Looking forward to productive discussions in Nashville! 🎯

[Ali Vosoughi @ali_vosough]

@ZacharyHuang12 Congratulations! 🎉🎉🎉

[Ali Vosoughi @ali_vosough]

@HosseinSouri8 Congrats! 🎉

[Ali Vosoughi @ali_vosough]

Exploring new platforms as @X evolves! Join me on @bluesky to reconnect and build a vibrant community. Excited for this fresh start!
Find me on Bluesky: https://t.co/wxWmXD80wl

[Ali Vosoughi @ali_vosough]

@msedebrahimi @PinterestEng Congrats🎉🎉