Mohit Bansal
@mohitban47
Followers
11K
Following
15K
Media
116
Statuses
5K
Parker Distinguished Prof @UNC. PECASE/AAAI Fellow. Director https://t.co/5qlPVgnrlN (@unc_ai_group). Past @Berkeley_AI @TTIC_Connect @IITKanpur #NLP #CV #AI
Joined September 2012
@cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin @ZiyangW00 @jaeh0ng_yoon @shoubin621 @mmiemon @gberta227 @pingzli @prateeky2806 @yilin_sung @TianlongChen4 @zhan1624 -- Video-Skill-CoT: Skill-based Chain-of-Thoughts for Domain-Adaptive Video Reasoning @danadaeun @jaeh0ng_yoon @jmin__cho
https://t.co/72zuoF1tbE
https://t.co/wzm9hR01Nh
arxiv.org
Recent advances in Chain-of-Thought (CoT) reasoning have improved complex video understanding, but existing methods often struggle to adapt to domain-specific skills (e.g., event detection,...
Excited to share Video-Skill-CoT๐ฌ๐ ๏ธโ a new framework for domain-adaptive video reasoning with skill-aware Chain-of-Thought (CoT) supervision! โก๏ธKey Highlights: โก๏ธ Automatically extracts domain-specific reasoning skills from questions and organizes them into a unified taxonomy,
0
1
3
@cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin @ZiyangW00 @jaeh0ng_yoon @shoubin621 @mmiemon @gberta227 @pingzli @prateeky2806 @yilin_sung @TianlongChen4 -- MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation @shoubin621 @zhan1624 @ZiyangW00 @jaeh0ng_yoon
https://t.co/j96f2ZPRCF
https://t.co/etZllmBg2H
arxiv.org
Combining pre-trained expert models offers substantial potential for scalable multimodal reasoning, but building a unified framework remains challenging due to the increasing diversity of input...
New paper Alert ๐จ Introducing MEXA: A general and training-free multimodal reasoning framework via dynamic multi-expert skill selection, aggregation and deep reasoning! MEXA: 1. Selects task- and modality-relevant experts based on the query and various required multimodal
1
2
6
@cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin @ZiyangW00 @jaeh0ng_yoon @shoubin621 @mmiemon @gberta227 -- Glider: Global and Local Instruction-Driven Expert Router @pingzli @prateeky2806 @jaeh0ng_yoon Jie Peng @yilin_sung @TianlongChen4
https://t.co/5s0jChd7u3
https://t.co/wcpA76pTcm
๐ Introducing GLIDER: Global and Local Instruction-Driven Expert Router! Our new approach combines LLM-generated semantic task instructions for global task-level routing with learned local token-level routing for improved performance on both held-in and held-out tasks. 1๏ธโฃ
1
0
2
@cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin @ZiyangW00 @jaeh0ng_yoon @shoubin621 @mmiemon @gberta227 -- RACCooN: A Versatile Instructional Video Editing Framework with Auto-Generated Narratives @jaeh0ng_yoon @shoubin621
https://t.co/Wx3eIJKDfp
https://t.co/2XV9Yhjx9l
arxiv.org
Recent video generative models primarily rely on carefully written text prompts for specific tasks, like inpainting or style editing. They require labor-intensive textual descriptions for input...
๐จNew paper๐RACCooN: remove/add/change video content effortlessly/interactively via our MLLM+Video Diffusion (V2P2V) framework with auto-generated descriptions! โถ๏ธ 1. Video-to-Paragraph (V2P): RACCooN first generates well-structured/detailed descriptions of videos with MLLM
1
1
3
@cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin @ZiyangW00 @jaeh0ng_yoon @shoubin621 @mmiemon @gberta227 -- Language Models Identify Ambiguities and Exploit Loopholes Jio Choi @EliasEskin
https://t.co/woCiJdD1VT
https://t.co/U2d0BUGxTN
๐จ 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
1
0
3
@cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin -- Video-RTS: Rethinking Reinforcement Learning and Test-Time Scaling for Efficient and Enhanced Video Reasoning @ZiyangW00 @jaeh0ng_yoon @shoubin621 @mmiemon @gberta227
https://t.co/THxKAhgCPX
https://t.co/c6s8hnrKFH
๐จIntroducing Video-RTS: Resource-Efficient RL for Video Reasoning with Adaptive Video TTS! While RL-based video reasoning with LLMs has advanced, the reliance on large-scale SFT with extensive video data and long CoT annotations remains a major bottleneck. Video-RTS tackles
1
3
7
(detailed links/websites + summary ๐งต's of these papers attached below FYIย ๐) -- MAgICoRe: Multi-Agent, Iterative, Coarse-to-Fine Refinement for Reasoning @cyjustinchen @ArchikiPrasad @swarnaNLP @EliasEskin
https://t.co/i5PpdcSCSg
https://t.co/zRHwIplma1
Aggregation & refinement improve LLM reasoning, but aggregation saturates, while refinement has 3 issues: 1) over-correction for easy problems 2) fails to localize+fix its own errors 3) insufficient number of refinement iteration for hard problems ๐จMulti-Agent, Iterative,
1
1
4
FYI, info/tags of folks presenting at @emnlpmeeting --> in-person: @jaeh0ng_yoon, @shoubin621 virtual: @ZiyangW00 @cyjustinchen @EliasEskin @danadaeun
0
3
9
๐จ Check out our awesome students/postdocs' papers at #EMNLP2025 and say hi to them ๐! Also, I will give a keynote (virtually) on "Attributable, Conflict-Robust, and Multimodal Summarization with Multi-Source Retrieval" at the NewSumm workshop. -- Jaehong (in-person) finished
2
28
62
๐จ Excited to announce Gistify!, where a coding agent must extract the gist of a repository: generate a single, executable, and self-contained file that faithfully reproduces the behavior of a given command (e.g., a test or entrypoint). โ
It is a lightweight, broadly applicable
2
37
90
๐ฅThe deadline (Nov 3, 2025 AoE) for ๐๐๐ฎ๐ซ๐๐๐ ๐๐๐๐ ๐๐จ๐ซ๐ค๐ฌ๐ก๐จ๐ฉ ๐จ๐ง ๐๐จ๐๐ข๐๐ฅ๐ฅ๐ฒ ๐๐๐ฌ๐ฉ๐จ๐ง๐ฌ๐ข๐๐ฅ๐ ๐๐ง๐ ๐๐ซ๐ฎ๐ฌ๐ญ๐ฐ๐จ๐ซ๐ญ๐ก๐ฒ ๐
๐จ๐ฎ๐ง๐๐๐ญ๐ข๐จ๐ง ๐๐จ๐๐๐ฅ๐ฌ (๐๐๐ฌ๐ฉ๐จ๐ง๐ฌ๐ข๐๐ฅ๐๐
๐) is approaching!๐ฅ ๐ Hybrid (Hilton Mexico City Reforma +
0
16
36
๐จ Proud to share our #TACL work on localizing factual inconsistencies in attributable text generation! To find where LLMs hallucinate, we need to get granular. We introduce QASemConsistency, a new method that decomposes text into simple question-answer pairs to precisely
0
12
23
๐ Excited to share that 5/5 of my papers (3 main, 2 findings) have been accepted at #EMNLP2025, in video/multimodal reasoning, instructional video editing, and efficient LLM adaptation & reasoning! ๐จ Iโm recruiting Ph.D. students to join the Multimodal AI Group at NTU College
15
31
304
Social dinner + gala in a beautiful 800-year old Bologna palace (Palazzo Re Enzo) right next to the famous Neptune's Fountain ๐
0
2
10
More info: https://t.co/hKvH2k1BLF Some other interesting facts: -- ECAI started in and has been running since 1974. -- This year it was held at the University of Bologna, which is the oldest university in continuous operation in the world, and the first degree-awarding
1
1
6
It was an honor and pleasure to give a keynote at the 28th European Conference on Artificial Intelligence (#ECAI2025) in beautiful Bologna, and engage in enthusiastic discussions about trustworthy + calibrated agents, collaborative reasoning + privacy, and controllable multimodal
1
26
68
Check out her work here โ https://t.co/r1M8XYFiJZ Google announcement blog (congrats to all the other fellows too) โ
blog.google
Today, we are announcing the recipients of the 2025 Google PhD Fellowship Program.
0
0
5
๐ Big congratulations to Vaidehi on being awarded a Google PhD Fellowship in Machine Learning and ML Foundations for her important research contributions in machine unlearning for LLMs/VLMs, defenses against adversarial attacks, and multi-agent privacy! #ProudAdvisor ๐๐
๐ฅณ๐ฅณ Honored and grateful to be awarded a 2025 Google PhD Fellowship in Machine Learning and ML Foundations for my research on machine unlearning, defenses against adversarial attacks, and multi-agent privacy! โจ Deep gratitude to my advisor @mohitban47 for his constant
3
13
143
I would encourage technical AI types to consider working in grantmaking! Schmidt Sciences is hiring for a unique position where you get to continue your own research at the same time Link:
jobs.lever.co
Summary Schmidt Sciences invites recent PhD graduates in AI and computer science to apply for a 12-18 month fellows-in-residence program. Reporting to the Director of the AI Institute at Schmidt...
4
29
145