Mohit Bansal

News:

(01/26) 12 new publications in ICLR, AAAI, EACL, TMLR on unguided world model inference, nudging LLM reasoning with hints, scene graph guided LLMs-as-a-Judge, codebase-level understanding via runtime execution, vision expert transformer for robot learning, data recipes for reasoning models, dynamic benchmarking on trustworthiness, evaluating MLLMs on image rotation, multi-agent disagreement, instruction tuning with and without context, video generation with retrieval-augmented motion adaptation, and language-based video reasoning.
(12/25) Honored and humbled to be selected as a ACL Fellow -- all the credit belongs to my amazing past and current students+postdocs+collaborators and mentors+family!
(12/25) Associate Editor-in-Chief for IEEE TPAMI journal -- we welcome everyone to submit their papers!
(10/25) Congrats to Vaidehi Patil for the Google PhD Fellowship!
(08/25) 14 new publications in NeurIPS, EMNLP, TMLR, TACL on bridging MMLM+diffusion models, multi-armed bandit reward selection, 4D space-time reconstruction, reward-driven video understanding, coarse-to-fine reasoning refinement, RL and TTS for video reasoning, loophole exploitation in LLMs, editing videos via auto-generated narratives, global+local instruction-driven expert router, general multimodal reasoning with dynamic multi-Expert aggregation, skill-based CoT for domain-adaptive video reasoning, math reasoning via fine-grained data synthesis analysis, localizing factual inconsistencies, and reliable+responsible foundation models.
(06/25) Congrats for job placements to postdocs/PhD students: Elias Stengel-Eskin: UT Austin CS faculty; Jaehong Yoon: NTU Singapore faculty; Jaemin Cho: JHU CS faculty; Jialu Li: Adobe Research Scientist; Yi-Lin Sung: Meta Research Scientist; Prateek Yadav: Meta Research Scientist.
(06/25) Congrats to Zaid Khan for the NDSEG PhD Fellowship!
(06/25) 17 new publications in COLM, ICCV, ACL, ICML, CVPR, TMLR, TACL on unit test generation, RAG conflict, task-circuit quantization, attribution generation programs, fine-grained summarization evaluation, occlusion understanding in VLMs, instructional video editing/reasoning, state-adaptive MoE-based VL-navigation, model merging at scale, multi-attribute steering of LLMs, localized attribution in content-grounded generation, self-consistency preference optimization, tree-based representation for long video reasoning, motion-grounded video reasoning, factual diverse generation via asymptotic entropy, model MoErging, and compression for communicating parameter-efficient updates.
(03/25) Congrats to Archiki Prasad for the Apple AI/ML PhD Fellowship!
(01/25) 11 new publications in ICLR 2025 (incl. a top-1.8% 'oral' and top-5% 'spotlight') on data generation environments/agents/policies, adapting diverse controls to diffusion models, balancing fast and slow planning, multimodal compositional+modular video reasoning, lifelong multimodal instruction tuning via dynamic data selection, safe T2I/T2V generation, generative infinite games, procedural+predictive video representation learning, bootstrapping language-guided navigation learning, automated preference data synthesis, and diagnosing cultural bias of VLMs.
(01/25) 5 new publications in NAACL 2025 on balancing agents' persuasion resistance+acceptance, adaptive decoding to balance contextual+parametric knowledge conflicts, reverse thinking for stronger LLM reasoning, positional bias of faithfulness for long-form summarization, and improving generation faithfulness via multi-agent collaboration.
(01/25) Honored and humbled to be selected as a AAAI Fellow -- all the credit belongs to my amazing past and current students+postdocs+collaborators and mentors+family!
(01/25) Honored and humbled to receive the Presidential Early Career Award for Scientists and Engineers (PECASE) Award by the US White House and President (was nominated in 2019, and awarded ARO-ECASE funding in 2021 as a `bridge program' due to White House PECASE announcement delays).
(11/24) Congrats to David Wan for the Google PhD Fellowship!
(11/24) 7 new publications in NeurIPS 2024 and TMLR on agent confidence calibration via speaker-listener pragmatics, merging skill-specific text-to-image experts, game-theoretic LLM evaluation, model editing and rational belief revision, unlearning in multimodal LLMs, vision-language-navigation survey, and efficient multimodal generation.
(07/24) 4 new publications in COLM 2024 and ECCV 2024 on adaptive/dynamic environment generation, long multi-scene consistent and fine-grained layout control for video generation, diagram generation via LLM planning, and contrastive region guidance for VLM grounding.
(06/24) Congrats for job placements to PhD students: Peter Hase: Anthropic Residency; Yichen Jiang: Apple AI/ML Research Scientist; Adyasha Maharana: Databricks/Mosaic Research Scientist; Swarnadeep Saha: FAIR-Meta Research Scientist.
(05/24) 12 new publications in ICML 2024 and ACL 2024 on multi-agent reasoning collaboration, inducing systematicity in transformers, easy-to=hard generalization, very long conversational memory of LLM agents, soft self-consistency, fine-grained hallucination evaluation, structured distillation of multi-agent reasoning interaction graphs, refactoring programs to discover abstractions, etc.
(04/24) Congrats to Jaemin Cho for the Bloomberg PhD Fellowship!
(04/24) Program Co-chairing for EMNLP 2024. We welcome everyone to submit their papers. Looking forward to welcoming you in Miami Nov12-16!
(03/24) 11 new publications in CVPR 2024 (incl. a 'spotlight') and NAACL 2024 and TMLR 2024 and EACL 2024 on interleaved+interactive any-to-any generation, alternating unimodal adaptation, interactive image segmentation, branch-solve-merge for LLM eval+generation, as-needed decomposition and planning with LLM agents, aspect-controlled referring expression generation, model merging by task subspace matching, low-cost algorithmic recourse, unified embeddings for multimodal retrieval, and hierarchical+dynamic prompt compression.
(01/24) 7 new publications (incl. 2 'spotlights') in ICLR 2024 on attack vs. defense for sensitive information deletion in LLMs, Davidsonian Semantics Scene Graphs for reliable T2I evaluation, data pruning with balanced diversity and difficulty, rephrase-based visual question grounding for VLMs, efficient sparse mixture-of-experts, efficient coarse-to-fine pruning for VLMs, analyzing and mitigating hallucinations in VLMs.
(10/23) Honored and humbled to receive the IIT Kanpur Young Alumnus Award.
(10/23) 6 new publications in EMNLP 2023 on reasoning chain/chain-of-thoughts evaluation, context dependency in language generation, data factors for compositional generalization, summary generation with controllable readability levels, multimodal model merging, and causal debiasing of multimodal models.
(09/23) 9 new publications (including 2 'spotlights') in NeurIPS 2023 on visual programming for interpretable+explainable text-to-image generation and evaluation, any-to-any multimodal generation, LLMs teaching student models, panorama generation for VLN generalization, model merging with interference, localization informing model editing, self-chained videoQA and localization, action knowledge in video-language LLMs, adaptive contextual perception.
(07/23) 3 new publications (including 1 'oral') in ICCV 2023 on text-to-image skill+bias evaluation, scaling VLN, and unified coarse-to-fine alignment for video-text retrieval.
(06/23) Postdocs Elias Stengel-Eskin and Jaehong Yoon joining us!
(05/23) 11 new papers in ACL 2023 and IJCAI 2023 and *SEM 2023 on faithful extractive summarization, meeting QA, single-frame bias in VidL, mixed fwd/rev cross-entropy, multimodal graph script induction, continual learning for code generation, exclusive supermasks for CL, modular multi-step reasoning, factuality of LLMs, sequential instruction understanding, and compositional LM differentiable prompting.
(02/23) 13 new papers in CVPR 2023 and ICLR 2023 and EACL 2023 and WACV 2023 on unified vision-text-layout DocAI, VLN with future-view semantics, hierarchical retrieval and step captioning, video-language pretraining, parameter-efficient audio-visual adapters, summarization programs, faithful decoding, edit-based prompt search, model belief graphs, social commonsense, multi-doc summarization + graph IE, long-distance video QA, perceiver-VL.
(12/22) Congrats to Zineng Tang for being selected as winner of the CRA Outstanding Undergraduate Researcher Award 2023!
(10/22) New papers in EMNLP 2022 and COLING 2022 and TACL on multimodal summarization + factuality, explanation hardness, mutual exclusivity for compositionality, action learning in interactive visual environments, LM bias, multimodal coreference in multi-turn dialogue, graph generative commonsense reasoning, and survey on data augmentation for limited data learning.
(09/22) Congrats to Swarnadeep Saha for the Google PhD Fellowship!
(09/22) New papers (including 3 selected/featured 'orals') in NeurIPS 2022 and ECCV 2022 on textless VL transformers, visual feature importance, ladder side-tuning, LLMs as fewshot video learners, T-few/IA3, gamified VL association benchmark, text to visual story generation, efficient video retrieval via audio replacement, etc.
(06/22) New papers in NAACL 2022 on factual summarization pretraining, factuality evaluation, scene imagination commonsense, multilingual vision-lang-navigation, fine-grained CLIP captioning, masked POS modeling, curriculum learning, video intent discovery, multi-doc summarization clustering, interactive summarization, etc.
(05/22) Promoted to full professor.
(03/22) Congrats to Yichen Jiang for the Apple AI/ML PhD Fellowship!
(03/22) New papers in CVPR 2022 and ACL 2022 on environment image editing for VLN generalization, vision-language parameter-efficient adapters, Cherokee revitalization NLP roadmap, explanation graph contrastive learning, and predicting human opinion distributions for NLI.
(12/21) Congrats to Shiyue Zhang for the Bloomberg PhD Fellowship!
(11/21) Congrats to Ori+Ram and team on the CoNLL 2021 Best Paper Runner-Up Award!
(09/21) New papers in NeurIPS 2021 on video-to-language knowledge distillation, query-based video highlight/saliency, socially-aligned feature importance explanations, video-lang understanding multi-task benchmark.
(09/21) Congrats to Peter Hase for the Google PhD Fellowship! (article)
(08/21) New papers in EMNLP 2021 on structure/compositionality, generalization, evaluation, and efficiency for explainability, commonsense, navigation, story visualization, and summarization.
(07/21) Thanks to NSF for funding our ENGAGE NSF-AI Institute! (article)
(06/21) Congrats to Jie and co-authors on the CVPR 2021 Best Student Paper Honorable Mention!
(05/21) New papers in ICML 2021 and ACL 2021 on unified vision-language generation, email thread summarization, continuous flow generation, multilingual video retrieval, dialogue-contradictions, cross-modal fake news detection, code generation, cherokee interactive demo, and chat disentaglement.
(04/21) Congrats to Xiang and co-authors on the EACL 2021 Best Long Paper Award Honorable Mention!
(04/21) New papers in NAACL 2021 on multi-proof generation, consistent visual story generation, localized video-language pretraining, syntactic vision-language navigation, dynamic benchmarking, tensor-product summarization, graph-based multi-doc summarization, and interactive multi-doc summarization, and robustness gym.
(03/21) Congrats to Jie Lei for the Adobe Research Fellowship!
(03/21) Honored and humbled to receive the Early Career Award for Scientists and Engineers (ECASE) by ARO!
(02/21) New papers in AAAI 2021, EACL 2021, and CVPR 2021 on pose-correction captioning, query-focused multi-doc summarization, commonsense graph-task matching, unreliable news detection, and efficient video-language pretraining.
(12/20) Hao's work on vokenization covered in MIT Technology Review.
(09/20) New papers in EMNLP 2020 on datasets (e.g., Cherokee MT, navigation+assembly, conjunctive/distributional NLI, manyhop-fact-verification) and methods (e.g., vokenization, proof generation, leakage-simulatability).
(08/20) Thanks to UNC for the UNC Phillip & Ruth Hettleman Prize for Artistic and Scholarly Achievement.
(05/20) Thanks to IJCAI for the `IJCAI Early Career Spotlight'.
(04/20) 8 new papers (6 in ACL 2020, 1 in IJCAI 2020, and 1 in ECCV 2020).
(02/20) Thanks to Amazon for the Amazon Machine Learning Research Award (blog post).
(01/20) Thanks to DARPA for the DARPA Director's Fellowship.
(01/20) Thanks to Microsoft for the Microsoft Investigator Fellowship (MSR blog post).
(12/19) Congrats to Sweta Karlekar and Han Guo for winning the Runner-Up and Finalist positions in the CRA Outstanding Undergraduate Researcher Awards! (link)
(11/19) 5 new papers in AAAI 2020 and ICRA 2020.
(08/19) Congrats to Peter Hase for the Royster Society PhD Fellowship!
(08/19) 5 new papers in EMNLP 2019.
(07/19) Thanks to NSF for the NSF-CAREER Award (details).
(07/19) Thanks to Google AI for the Google Focused Research Award (details).
(06/19) Congrats to Hao Tan for the Bloomberg Data Science Ph.D. Fellowship!
(05/19) 6 new papers in ACL 2019 (congrats to Hyounghun for best paper nomination) and 2 new preprints.
(04/19) Congrats to Darryl Hannan for the 3-year NSF PhD Fellowship!
(02/19) 5 new papers: 3 in NAACL 2019, 1 in CVPR 2019, 1 in ICRA 2019.
(01/19) Congrats to Ramakanth Pasunuru for the 2-year Microsoft Research PhD Fellowship (and finalist for Facebook PhD Fellowship)!
(12/18) Congrats to Han Guo for the CRA Outstanding Undergraduate Researcher Award Honorable Mention!
(11/18) Thanks to research awards from Salesforce, Facebook, and IBM.
(10/18) 2 new papers in AAAI 2019 (16% acceptance rate).
(08/18) 7 new papers (6 in EMNLP; 1 in CoNLL -- see below).
(07/18) Thanks to Army Research Office for the ARO Young Investigator Program (YIP) Award.
(07/18) 1st rank in EMNLP FEVER (Fact Extraction & VERification) shared task (congrats Yixin, Haonan)! [Press Article]
(06/18) COLING paper on dynamic-MTL selected as "Area Chair Favorites" (congrats Han+Ram)!
(04/18) 4 new papers (2 in ACL; 1 in TACL; 1 in WiNLP -- see below).
(04/18) Congrats to Lisa Bauer for the 3-year NSF PhD Fellowship!
(03/18) Thanks to Adobe for the Adobe Research Award.
(02/18) 9 new 2018 papers in NAACL, CVPR, AAAI, WACV (see below).
(09/17) Thanks to DARPA for the DARPA Young Faculty Award (link).
(09/17) Thanks to Facebook for the Facebook ParlAI Research Award.
(06/17) Top single model results on the RepEval-NLI Shared Task at EMNLP 2017 (congrats Yixin!).
(06/17) Outstanding Paper Award at ACL 2017 (congrats Ram!).
(02/17) Thanks to Google for a Google Faculty Research Award (link).
(07/16) Best paper award at ACL 2016 Repl4NLP workshop for paper on mapping unseen words.
(03/16) Thanks to Bloomberg for a Bloomberg Data Science Research Grant (link).
(02/16) Paper on universal sentence embeddings selected as an oral at ICLR 2016.
(01/16) Our work on AI for computational humor was covered in MIT Technology Review and Newsweek.
(11/15) Nvidia paper award at NIPS 2015 Multimodal ML workshop for paper on navigational instruction following.
(12/14) Thanks for an IBM Faculty Award and a Google Faculty Research Award (link).

Computer Science department, UNC Chapel Hill (2016 — present) John R. & Louise S. Parker Distinguished Professorship Director, MURGe-Lab (UNC-AI Group) Full Professor 2022-present Associate Professor: 2020-2022 Assistant Professor: 2016-2020
Toyota Technological Institute, Chicago (2013 — 2016) Research Assistant Professor
EECS, UC Berkeley (2008 — 2013) Graduate Student Researcher (Advisor: Dan Klein)
Google Research, Mountain View (Summer 2011) Research Intern (with John DeNero and Dekang Lin)
Microsoft Research, Redmond (Summer 2010) Research Intern (with Chris Quirk and Bob Moore)
Cornell University, Ithaca (Summer 2007) Research Intern (with Lillian Lee and Claire Cardie)

Mohit Bansal

Parker Distinguished Professor

Recent Invited Talks/Keynotes:

News:

About

Work Experience

Selected Awards/Honors

Publications (+Code/Data)

Talks/Teaching

Students/Postdocs

Professional Service