@wregss

Student Researcher @GoogleDeepMind PhD-ing @WisconsinCS | MS at @uwcse/@RAIVNLab | Ex @AIatMeta @nvidiaai and @samsungresearch. He/Him

 Madison, WI
aniketrege.github.io
Joined June 2009
  • Tweets1.2K
  • Following469
  • Followers1.1K
163Media
🎉 Super stoked to share that our work is accepted to the main conference of ACL 2026!! See you in sunny San Diego 🌞 #ACL2026 #NLProc Paper thread below 🧵
1
4
54
2,462
Q for all RL post training experts from a noob - what are some good detailed resources about building data mixtures to improve "small"-ish models (<70B) on specific goals/skills e.g. multi-hop retrieval? Tulu 3 (@natolambert et al) has been the best I've found so far!
1
9
1,710
Folks at #acl2026 interested in long-horizon agentic reasoning, this is happening now in Grand Hall!!
Ramya Vinayak@ramyavinayakJul 3
How can AI understand hours of first-person video instead of just a few short clips? At #ACL2026 we’re presenting EGAgent, a framework for long-horizon video understanding led by my student @wregss 🧵[1/4]
5
326
Aniket Rege retweeted
Ramya Vinayak@ramyavinayakJul 2
Excited to be at #ACL2026. Delighted to present keynote talk tomorrow on learning personalized preferences at 📍 Workshop on Customizable NLP @customnlp4u 🗓️ Friday, July 3rd | ⏰ 9:10–9:50 am PT 📌Old Town
2
7
26
1,215
🚨New Paper🚨 Learning to Plan and Orchestrate for Open-Ended Image Editing Image editors handle concrete tasks, but struggle with abstract ones (e.g., adapt this ad for winter), which require coordinated edits. We tackle this via planning and experiential tool learning. 🧵👇
1
2
4
313
we both up now
Zohran Kwame Mamdani
@ZohranKMamdani22 Nov 2020
never thought I’d be more hopeful about socialism than arsenal and yet here we are.
785
24,318
911
308,178
11,351,606
🚨 Very happy to have advised this #icml2026 work led by @MaheshRam23629 exploring RL for cooperative reasoning in LLMs tl;dr post training LLMs to play the card game Hanabi 🎆 improves performance on unseen tasks like instruction following and temporal reasoning!
Mahesh Ram@MaheshRam23629May 19
🔥Our paper “Sparks of Cooperative Reasoning: LLMs as Strategic Hanabi Agents” was accepted to ICML 2026! We show that post-training LLMs with RL to play a cooperative game teach generalizable cooperative reasoning. 1/n
12
1,488
😭😭
rohit
@krishnanrohitMay 14
In hindsight Shl0ms really should've disclosed upfront that the painting was done by Claude
1
301
Hanging around @Stanford / Palo Alto for the rest of the day, any mutuals who are free for a coffee or chat pls DM! 🤓
4
391
Hi #ml twitter! I’m in San Francisco from 5/6-5/12, I’d love to catch up over a coffee and chat about research and things y’all are working on!! (post @NeurIPSConf submission hangover 🤓)
20
1,463
Aniket Rege retweeted
Jaden Park
@_jadenparkApr 16
We all knew LLM agents struggle to explore, but we had to eyeball it 👀. We couldn't measure exploration errors. Until now. 🗺️🤖 We built a policy-agnostic metric to quantify exploration and exploitation errors in LLM agents. Spoiler: Exploration error is what kills📉 agent performance in our setting 👇🧵(1/8)
1
17
31
2,352
Excited to see Qwen-3.5-Omni evaluated on AV-SpeakerBench 🚀 From ~54 → ~71.5: open models are catching up in audiovisual reasoning. Gemini still leads, but the gap is shrinking. Special thanks to my mentor, @yu_zhuoran32720 , and my advisor, @yong_jae_lee , from the Wisconsin AI Vision Lab, for making AV-SpeakerBench possible. Two questions: - "Is Qwen-3.5-Omni really open-source?" - "Why no strong AV models yet from @OpenAI @AnthropicAI ?" AV reasoning is the next frontier. 🔗 Benchmark Project Page: plnguyen2908.github.io/AV-Sp… 🔗 Qwen blog: qwen.ai/blog?id=qwen3.5-omni…
Qwen
@Alibaba_QwenMar 30
🚀 Qwen3.5-Omni is here! Scaling up to a native omni-modal AGI. Meet the next generation of Qwen, designed for native text, image, audio, and video understanding, with major advances in both intelligence and real-time interaction. A standout feature: 'Audio-Visual Vibe Coding'. Describe your vision to the camera, and Qwen3.5-Omni-Plus instantly builds a functional website or game for you. Offline Highlights: 🎬 Script-Level Captioning: Generate detailed video scripts with timestamps, scene cuts & speaker mapping. 🏆 SOTA Performance: Outperform Gemini-3.1 Pro in audio and matches its audio-visual understanding. 🧠 Massive Capacity: Natively handle up to 10h of audio or 400s of 720p video, trained on 100M+ hours of data. 🌍 Global Reach: Recognize 113 languages (speech) & speaks 36. Real-time Features: 🎙️ Fine-Grained Voice Control: Adjust emotion, pace, and volume in real-time. 🔍 Built-in Web Search & complex function calling. 👤 Voice Cloning: Customize your AI's voice from a short sample, with engineering rollout coming soon. 💬 Human-like Conversation: Smart turn-taking that understands real intent and ignores noise. The Qwen3.5-Omni family includes Plus, Flash, and Light variants. Try it out: Blog: qwen.ai/blog?id=qwen3.5-omni Realtime Interaction: click the VoiceChat/VideoChat button (bottom-right): chat.qwen.ai HF-Demo: huggingface.co/spaces/Qwen/Q… HF-VoiceOnline-Demo: huggingface.co/spaces/Qwen/Q… API-Offline: alibabacloud.com/help/en/mod… API-Realtime: alibabacloud.com/help/en/mod…
3
7
1,168
🚨 I'm super stoked to share that I'll be spending this summer as a student researcher at the GenAI team of @GoogleDeepMind with @_gaganm , working on long horizon tool use!!
5
4
102
5,278
A HUGE shoutout to my network for your overwhelming response to my internship search post :') I had SO many wonderful conversations and I'm so thankful for all your generosity and kindness! xitter.cf/wregss/status/20336367…
Aniket Rege@wregssMar 16
Hi ML Twitter! My Summer 2026 internship unfortunately fell through last minute 😵‍💫 If your team is looking for interns, I’d love to connect - RTs appreciated 🙏 My website: aniketrege.github.io/
3
502
📢 Reminder: The submission deadline for the Machine Translation for Vision (MTV) Challenge at MAPS @CVPR2026 has been extended to April 17th!
1
2
3
770
xitter.cf/wregss/status/20346774…
Aniket Rege@wregssMar 19
🚨New work with @Meta @RealityLabs We introduce EGAgent, an agentic reasoning framework for very long video understanding powered by entity scene graphs Why? With long multimodal data streams, agents must search and reason across multiple modalities! 🧵 (1/n)
1
2
206
If you’re working on benchmarks for cultural authenticity or methodology for AI pluralism, we want to see your work at the MAPS Workshop @CVPR 2026. We are also hosting the Machine Translation for Vision (MTV) challenge, don't miss out! 🏆 Check out the full CFP below. 👇
MAPS - CVPR 2026 Workshop@maps_cvprMar 19
📢 Submissions are now OPEN for our @CVPR Workshop: Multimodal Alignment for a Pluralistic Society (MAPS)! Help us build AI that reflects global diversity, culture, and human values. 🌍🌏🌎 📅 Mar 3 – Apr 10, 2026 📝 Short papers (4 pgs) 🔗 openreview.net/group?id=thec… #CVPR2026
2
6
427
Hi computer vision fam! Interested in generative models that suit a diverse plurality of human values and cultures? Please submit to the @maps_cvpr workshop at #CVPR2026 led by the amazing @PShravannayak which I'm helping co-organize! Looking forward to your exciting works!!
MAPS - CVPR 2026 Workshop@maps_cvprMar 19
📢 Submissions are now OPEN for our @CVPR Workshop: Multimodal Alignment for a Pluralistic Society (MAPS)! Help us build AI that reflects global diversity, culture, and human values. 🌍🌏🌎 📅 Mar 3 – Apr 10, 2026 📝 Short papers (4 pgs) 🔗 openreview.net/group?id=thec… #CVPR2026
1
1
393
Huge gains in Long-Horizon Relational Reasoning! On tasks that require modeling relationships between entities across long time horizons, EGAgent excels compared to previous SoTA (EgoButler) and Gemini 2.5 Pro 🔺 + 20.8 % on RelationMap 🔺 + 22.2 % on TaskMaster 🧵(6/n)
1
1
175
📹 Not just EgoCentric Videos! On Video-MME (Long), we match strong 7B baseline AdaVideoRAG (Xue et al., NeurIPS ‘25) using Qwen-2.5 VL while processing over 10x fewer frames. Our framework is straightforward to adapt to any long horizon multimodal dataset! 🧵(7/n)
1
1
153
Joint work with amazing collaborators @AIatMeta @RealityLabs @UWMadison @ArkaSadhu29 @oi02lyl @kejie_li @ramyavinayak @yuning_chai @yong_jae_lee @kim_hyojin_ Project Page: facebookresearch.github.io/e… Code: github.com/facebookresearch/… ArXiv: arxiv.org/abs/2601.18157 🧵(8/n)
2
143
Load more