VideoDeepResearch: Long Video Understanding With Agentic Tool Using Overview: VideoDeepResearch presents an agentic framework for long video understanding, bypassing the need for extended-context MLLMs by combining a text-only large reasoning model with a modular multi-modal toolkit. This system generates problem-solving strategies through reasoning and selective tool-based content access, yielding performance increases of 9.6% on MLVU (test), 6.6% on LVBench, and 3.9% on LongVideoBench over prior MLLM baselines. Paper: arxiv.org/abs/2506.10821

🚨This week's top AI/ML research papers: - Self-Adapting Language Models - V-JEPA 2 - The Illusion of the Illusion of Thinking - Magistral - Reinforcement Pre-Training - VideoDeepResearch - Unsupervised Elicitation of LMs - CoRT - The Diffusion Duality - Ming-Omni - One Tokenizer To Rule Them All - Build the web for agents, not agents for the web - Learning What RL Can't - Hidden in plain sight - ViTs Don't Need Trained Registers - Play to Generalize - Thinking vs. Doing - Self Forcing - Highly Compressed Tokenizer Can Generate Without Training - RiemannFormer - Seedance 1.0 - Edit Flows - Resa - On a few pitfalls in KL divergence gradient estimation for RL - e3 overview for each authors' explanations read this in thread mode for the best experience

13. VideoDeepResearch: Long Video Understanding With Agentic Tool Using 🔑 Keywords: Long video understanding, multi-modal large language models, VideoDeepResearch, agentic systems, reasoning model 💡 Category: Multi-Modal Learning 🌟 Research Objective: To challenge the common assumption that long video understanding (LVU) requires multi-modal large language models (MLLMs) with extended context windows and specialized capabilities. 🛠️ Research Methods: Introduced VideoDeepResearch, a framework utilizing a text-only large reasoning model (LRM) paired with a modular multi-modal toolkit to approach LVU tasks through reasoning and selective content access. 💬 Research Conclusions: - VideoDeepResearch achieved notable improvements over existing MLLM baselines, with advances of 9.6%, 6.6%, and 3.9% on key benchmarks (MLVU, LVBench, LongVideoBench). - The results indicate that agentic systems hold significant potential for addressing challenges in long video understanding. 👉 Paper link: huggingface.co/papers/2506.1…

📚 AI Native Daily Paper Digest - 20250613 🌟 Follow @AINativeF for the latest insights on AI Native. Covering AI research papers from Hugging Face, featured in the image. 💡 Stay updated with the latest research trends and dive deep into the future of AI! 🚀 #AI #HuggingFace #AIPaper #AINative #AINF — Appendix: Today's AI research papers — 1. ReasonMed: A 370K Multi-Agent Generated Dataset for Advancing Medical Reasoning 2. SWE-Factory: Your Automated Factory for Issue Resolution Training Data and Evaluation Benchmarks 3. Text-Aware Image Restoration with Diffusion Models 4. VRBench: A Benchmark for Multi-Step Reasoning in Long Narrative Videos 5. AniMaker: Automated Multi-Agent Animated Storytelling with MCTS-Driven Clip Generation 6. Magistral 7. Discrete Audio Tokens: More Than a Survey! 8. Domain2Vec: Vectorizing Datasets to Find the Optimal Data Mixture without Training 9. Optimus-3: Towards Generalist Multimodal Minecraft Agents with Scalable Task Experts 10. PosterCraft: Rethinking High-Quality Aesthetic Poster Generation in a Unified Framework 11. Resa: Transparent Reasoning Models via SAEs 12. AutoMind: Adaptive Knowledgeable Agent for Automated Data Science 13. VideoDeepResearch: Long Video Understanding With Agentic Tool Using

VideoDeepResearch: Long Video Understanding With Agentic Tool Using A modular agentic framework for long video understanding that combines a text-only reasoning model (DeepSeek-R1-0528) with specialized tools such as retrievers, perceivers, and extractors to outperform large MLLMs without relying on extended context windows. PAPER: arxiv.org/abs/2506.10821