What if a Transformer could render? Not text → image. But mesh → image — with global illumination. No rasterizers. No ray-tracers. Just a Transformer without per-scene training. RenderFormer does exactly that. #SIGGRAPH2025 🔗microsoft.github.io/renderformer

Dear MAGA friends, I have been worrying about STEM in the US a lot, because right now the Senate is writing new laws that cut 75% of the STEM budget in the US. Sorry for the long post, but the issue is really important, and I want to share what I know about it. The entire

I’m thrilled to share that I’ll be joining the University of North Carolina at Chapel Hill for my CS PhD this fall!! 🎓💙 UNC-Chapel Hill I’ll be working with the amazing Mohit Bansal at UNC NLP. Grateful to everyone who’s supported me, excited for this new chapter! 🚀

Excited to share that I’ll be joining UNC Computer Science and UNC NLP as a Postdoctoral Research Associate, working with the incredible Mohit Bansal! Can’t wait to collaborate with the amazing students and faculty there! 🎉 A huge thank you to my supervisor Reza Haffari, my colleagues at

Welcome to Chapel Hill, Minghao! Can't wait to see the exciting research you'll be doing here as a postdoc 🙂

Excited to share Video-Skill-CoT🎬🛠️– a new framework for domain-adaptive video reasoning with skill-aware Chain-of-Thought (CoT) supervision! ⚡️Key Highlights: ➡️ Automatically extracts domain-specific reasoning skills from questions and organizes them into a unified taxonomy,

Introducing Video-Skill-CoT 📽️ , a new framework for domain-adaptive video understanding with skill-specific chain-of-thought reasoning! ✅ Automatically discovers reasoning skills from video data ✅ Trains skill-specific expert modules with skill-specific CoT rationales ✅

🚨 New Release: Video-Skill-CoT! Domain-Adaptive, Skill-Based Video Reasoning💡 ✅ Automatically extracts domain-specific reasoning skills ✅ Generates tailored, skill-based CoT rationales ✅ Trains with skill-specific experts for stronger domain adaptation 🚀 Outperforms

🚨Check my amazing labmate's latest work 🎬 Video-Skill-CoT 🛠️, a powerful and elegant framework for domain-adaptive video reasoning with skill-aware CoT 🧠✨, achieving strong results across multiple tasks! 📊🔥

🚨 CLATTER treats entailment as a reasoning process, guiding models to follow concrete steps (decomposition, attribution/entailment, and aggregation). CLATTER improves hallucination detection via NLI, with gains on ClaimVerify, LFQA, and TofuEval especially on long-reasoning

This paper proposes VIDEO-SKILL-COT to improve domain adaptation using skill-aware Chain-of-Thought supervision and expert learning modules. Methods 🔧: → The framework automatically constructs skill-based Chain-of-Thought annotations by extracting skills from questions,

Excited to share our new work, CLaMR! 🚀 We tackle multimodal content retrieval by jointly considering video, speech, OCR, and metadata. CLaMR learns to dynamically pick the right modality for your query, boosting retrieval by 25 nDCG@10 over single modality retrieval! 🧐

Excited to announce CLaMR, our new retriever for multimodal documents! Strong performance improvements (+25 nDGC@10) compared to both multimodal and unimodal retrieval baselines. 🤝 CLaMR jointly encodes multiple modalities and selects the most relevant ones for each query. 🏋️‍♂️

How can a multimodal retriever accurately retrieve docs from massive online video content that spans multiple modalities? We introduce CLaMR, a contextualized late-interaction retriever that jointly encodes all modalities and dynamically selects those containing the relevant

Introducing CLaMR -- a late-interaction retriever for complex multimodal video content! 📽️📚 ➡️ Jointly encodes frames, speech, on-screen text, and metadata to answer diverse queries grounded across modalities ➡️ Trained with a new dataset we introduce, MultiVENT 2.0++, a

Excited to present VideoTree🌲 at #CVPR2025 Fri at 10:30AM! VideoTree improves long-video QA via smart sampling: -Query-adaptive: finds the parts of the video relevant to the query -Coarse-to-fine structure: structured hierarchically to sample granularly from relevant segments

Wow I missed this extra fancy ColBERT model. > A late-interaction retriever which jointly encodes/contextualizes information from many modalities, allowing for fine-grained matching between the query and implicitly finding the most relevant modality.

Welcome Jaewoo to the MURGe-Lab + UNC NLP + UNC Computer Science family & the beautiful Chapel Hill + Research Triangle area! 🎉 Looking forward to the exciting research and fun together in your PhD journey 💙

Welcome Jaewoo! Looking forward to seeing your exciting future research at UNC MURGe lab 😄

🚨 Want models to better utilize and ground on the provided knowledge? We introduce Context-INformed Grounding Supervision (CINGS)! Training LLM with CINGS significantly boosts grounding abilities in both text and vision-language models compared to standard instruction tuning.