Zaid Khan @codezakh

Shiny new building and good views from the new VirginiaTech campus in DC 😉 -- it was a pleasure to meet everyone and engage in exciting discussions about trustworthy agents, collaborative reasoning/privacy, and controllable multimodal generation -- thanks again @profnaren,…

keyboard_arrow_left Previous keyboard_arrow_right Next

Mohit Bansal @mohitban47

6 days ago

Shiny new building and good views from the new VirginiaTech campus in DC 😉 -- it was a pleasure to meet everyone and engage in exciting discussions about trustworthy agents, collaborative reasoning/privacy, and controllable multimodal generation -- thanks again @profnaren,… https://t.co/1HCyQmCFID

2 22 60 20K 3

Proud to have been part of the team behind Gaia2 and ARE! ARE = a gym/platform for scaling up LLM agent envs for evals & RL Gaia2 = a new benchmark for hard & practical agent tasks (search, execution, ambiguity, time, noise, & multi-agent) tinyurl.com/aregaia2

🎉 Excited to share that Bifrost-1 has been accepted to #NeurIPS2025! ☀️ Bridging MLLMs and diffusion into a unified multimodal understanding and generation model can be very costly to train. ✨ Bifrost-1 addresses this by leveraging patch-level CLIP latents that are natively…

Han Lin @hanlin_hl

2 months ago

🎉 Excited to share that Bifrost-1 has been accepted to #NeurIPS2025! ☀️ Bridging MLLMs and diffusion into a unified multimodal understanding and generation model can be very costly to train. ✨ Bifrost-1 addresses this by leveraging patch-level CLIP latents that are natively…

2 58 150 40K 69

Download Image

Looking forward to giving this Distinguished Lecture at Virginia Tech on Friday & meeting the awesome faculty+students there! 🙂 Will also discuss some of our new work in multi-agent compositional privacy risks+mitigation, ambiguity/loophole exploitation by LLMs, and bridging…

The Sanghani Center at Virginia Tech @SanghaniCtrVT

7 days ago

Looking forward to giving this Distinguished Lecture at Virginia Tech on Friday & meeting the awesome faculty+students there! 🙂 Will also discuss some of our new work in multi-agent compositional privacy risks+mitigation, ambiguity/loophole exploitation by LLMs, and bridging…

0 2 12 9K 2

Download Image

🎉 Thrilled to share that our paper LASeR on using multi-armed bandits to select reward models for LLM training has been accepted to #NeurIPS2025! Check the post for more paper details and camera-ready updates. See you in San Diego! 👇

Archiki Prasad @ArchikiPrasad

7 days ago

🎉 Thrilled to share that our paper LASeR on using multi-armed bandits to select reward models for LLM training has been accepted to #NeurIPS2025! Check the post for more paper details and camera-ready updates. See you in San Diego! 👇

6 41 169 31K 84

It's application season, and I'm sharing some of my past application materials: - Academic job market (written in Dec 2024) - PhD fellowship (written in Apr 2023) - PhD admission (written in Dec 2019) on my website (j-min.io)

🚨 The Sum Leaks More Than Its Parts --> (1) Multi-agent compositional privacy risk framework that models how seemingly innocuous outputs that seem safe to each agent alone can be combined across agents and time to infer private info. (2) Theory-of-Mind defense to reason about…

Vaidehi Patil @vaidehi_patil_

2 weeks ago

🚨 The Sum Leaks More Than Its Parts --> (1) Multi-agent compositional privacy risk framework that models how seemingly innocuous outputs that seem safe to each agent alone can be combined across agents and time to infer private info. (2) Theory-of-Mind defense to reason about…

4 42 89 21K 40

Download Image

🚨 We study new privacy risks that emerge as systems become increasingly multi-agent: seemingly innocuous queries can be composed across distributed/siloed agents to infer private info. One solution: reason about the adversary's motivations/state and communicate across the org…

Vaidehi Patil @vaidehi_patil_

2 weeks ago

🚨 We study new privacy risks that emerge as systems become increasingly multi-agent: seemingly innocuous queries can be composed across distributed/siloed agents to infer private info. One solution: reason about the adversary's motivations/state and communicate across the org…

4 42 89 21K 40

Download Image

🚨 In multi-agent LLM systems, outputs that seem safe to each agent alone can be combined to spill sensitive info! We define this compositional privacy leakage for distributed LLM agents and introduce Theory-of-Mind and multi-agent Collaborative Consensus defenses to mitigate it.…

🚨 Excited to share new work on LLMs and loopholes, accepted to #EMNLP2025 main! When models are faced with conflicting goals and ambiguous instructions that would let them exploit a loophole, many of the strongest models (Qwen, GPT4o, Claude, Gemini) do. This is a new risk and…

🎉 Excited to share that our Video-Skill-CoT paper has been accepted to #EMNLP2025 Findings! Video-Skill-CoT is a domain-adaptive video reasoning framework that automatically constructs skill-aware Chain-of-Thought (CoT) supervisions. It builds a shared skill taxonomy from…

Daeun Lee @danadaeun

4 months ago

🎉 Excited to share that our Video-Skill-CoT paper has been accepted to #EMNLP2025 Findings! Video-Skill-CoT is a domain-adaptive video reasoning framework that automatically constructs skill-aware Chain-of-Thought (CoT) supervisions. It builds a shared skill taxonomy from…

2 32 83 28K 29

Download Video

🎉Excited to share that our MEXA paper is accepted to #EMNLP2025 Findings! 🚀MEXA is a general, training-free multimodal reasoning framework that dynamically selects and aggregates experts/skills for deep, free-form reasoning, and is flexible & extensible to new…

Shoubin Yu @shoubin621

3 months ago

🎉Excited to share that our MEXA paper is accepted to #EMNLP2025 Findings! 🚀MEXA is a general, training-free multimodal reasoning framework that dynamically selects and aggregates experts/skills for deep, free-form reasoning, and is flexible & extensible to new…

2 32 70 18K 25

Download Image

🎉 RACCooN got accepted at #EMNLP2025 Main! 🚀 Our MLLM+Video Diffusion (Video-to-Paragraph-to-Video, V2P2V) framework enables effortless video editing w/ auto-generated descriptions, multi-granular pooling & mask planning. RACCooN Achieves +9.4%p human eval & 49.7%↓ FVD,…

Jaehong Yoon @jaeh0ng_yoon

a year ago

🎉 RACCooN got accepted at #EMNLP2025 Main! 🚀 Our MLLM+Video Diffusion (Video-to-Paragraph-to-Video, V2P2V) framework enables effortless video editing w/ auto-generated descriptions, multi-granular pooling & mask planning. RACCooN Achieves +9.4%p human eval & 49.7%↓ FVD,…

1 34 85 18K 21

Download Image

Excited to share that MAgICoRe has been accepted to #EMNLP2025 main! 🎉 Our work identifies 3 key challenges in LLM refinement for reasoning: 1) Over-correction on easy problems 2) Fail to localize and fix its own errors 3) Too few refinement iterations for harder problems…

Justin Chih-Yao Chen @cyjustinchen

a year ago

Excited to share that MAgICoRe has been accepted to #EMNLP2025 main! 🎉 Our work identifies 3 key challenges in LLM refinement for reasoning: 1) Over-correction on easy problems 2) Fail to localize and fix its own errors 3) Too few refinement iterations for harder problems…

3 77 246 48K 200

Download Image

🎉Our Video-RTS paper has been accepted at #EMNLP2025 Main!! We propose a novel video reasoning approach that combines data-efficient reinforcement learning (GRPO) with video-adaptive test-time scaling, improving reasoning performance while maintaining efficiency on multiple…

Ziyang Wang @ZiyangW00

3 months ago

🎉Our Video-RTS paper has been accepted at #EMNLP2025 Main!! We propose a novel video reasoning approach that combines data-efficient reinforcement learning (GRPO) with video-adaptive test-time scaling, improving reasoning performance while maintaining efficiency on multiple…

1 33 42 16K 9

Download Image

📢 Introducing RotBench, which tests whether SoTA MLLMs (e.g., GPT-5, GPT-4o, o3, Gemini-2.5-pro) can identify the rotation of input images (0°, 90°, 180°, and 270°). Even frontier MLLMs struggle at this spatial reasoning task that humans solve with >98% Acc. ➡️ Models struggle…

Bifrost-1 Bridging Multimodal LLMs and Diffusion Models with Patch-level CLIP Latents

🤔 Can we bridge MLLMs and diffusion models more natively and efficiently, by having MLLMs produce patch-level CLIP latents already aligned with their visual encoders, while fully preserving MLLM's visual reasoning capabilities? Introducing Bifrost-1: 🌈 > High-Fidelity…

🚀 I'm recruiting PhD students to join my lab (jaehong31.github.io) at NTU Singapore (@NTUsg), starting Spring 2026. If you're passionate about doing cutting-edge and high-impact research in multimodal AI, Trustworthy AI, continual learning, or video generation/reasoning,…

Jaehong Yoon @jaeh0ng_yoon

4 months ago

🚀 I'm recruiting PhD students to join my lab (jaehong31.github.io) at NTU Singapore (@NTUsg), starting Spring 2026. If you're passionate about doing cutting-edge and high-impact research in multimodal AI, Trustworthy AI, continual learning, or video generation/reasoning,…

30 31 229 51K 28

Download Image

🚀 We introduce GrAInS, a gradient-based attribution method for inference-time steering (of both LLMs & VLMs). ✅ Works for both LLMs (+13.2% on TruthfulQA) & VLMs (+8.1% win rate on SPA-VL). ✅ Preserves core abilities (<1% drop on MMLU/MMMU). LLMs & VLMs often fail because…