ML Research Hub
32.8K subscribers
5.61K photos
356 videos
24 files
6.07K links
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.

Admin: @HusseinSheikho || @Hussein_Sheikho
Download Telegram
UniG2U-Bench: Do Unified Models Advance Multimodal Understanding?

📝 Summary:
Unified multimodal models generally underperform specialized VLMs in generation-to-understanding tasks. However, they show consistent enhancements in spatial intelligence, visual illusions, and multi-round reasoning. This highlights the need for diverse training data to unlock their full potential.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03241
• PDF: https://arxiv.org/pdf/2603.03241
• Project Page: https://nssmd.github.io/unig2u.github.io/
• Github: https://github.com/nssmd/UniG2U

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
NOVA: Sparse Control, Dense Synthesis for Pair-Free Video Editing

📝 Summary:
NOVA is a novel unpaired video editing framework that uses sparse semantic guidance and dense synthesis to achieve high-fidelity editing with improved motion preservation and temporal coherence. AI-ge...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02802
• PDF: https://arxiv.org/pdf/2603.02802
• Github: https://github.com/WeChatCV/NovaEdit

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Kiwi-Edit: Versatile Video Editing via Instruction and Reference Guidance

📝 Summary:
A scalable data generation pipeline creates high-fidelity video editing training data, and a unified architecture enables improved instruction-following and reference fidelity in controllable video ed...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02175
• PDF: https://arxiv.org/pdf/2603.02175
• Project Page: https://showlab.github.io/Kiwi-Edit/
• Github: https://github.com/showlab/Kiwi-Edit

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Kling-MotionControl Technical Report

📝 Summary:
Kling-MotionControl is a DiT-based framework for character animation that combines heterogeneous motion representations, adaptive identity-agnostic learning, and advanced acceleration techniques to ac...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03160
• PDF: https://arxiv.org/pdf/2603.03160

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
HateMirage: An Explainable Multi-Dimensional Dataset for Decoding Faux Hate and Subtle Online Abuse

📝 Summary:
HateMirage is a new dataset designed to advance research on hate speech embedded in misinformation by providing multi-dimensional annotations for target, intent, and implication, offering a benchmark ...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02684
• PDF: https://arxiv.org/pdf/2603.02684
• Github: https://github.com/Sai-Kartheek-Reddy/HateMirage

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
SGDC: Structurally-Guided Dynamic Convolution for Medical Image Segmentation

📝 Summary:
Structure-Guided Dynamic Convolution enhances medical image segmentation by using explicit structural guidance to preserve fine-grained details lost through traditional average pooling methods. AI-gen...

🔹 Publication Date: Published on Feb 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.23496
• PDF: https://arxiv.org/pdf/2602.23496
• Github: https://github.com/solstice0621/SGDC

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Token Reduction via Local and Global Contexts Optimization for Efficient Video Large Language Models

📝 Summary:
AOT framework reduces video token redundancy through local-global optimal transport to preserve informative contexts while achieving efficient spatiotemporal compression in video large language models...

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.01400
• PDF: https://arxiv.org/pdf/2603.01400
• Project Page: https://tyroneli.github.io/AOT/

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Media is too big
VIEW IN TELEGRAM
Track4World: Feedforward World-centric Dense 3D Tracking of All Pixels

📝 Summary:
A feedforward model called Track4World enables efficient holistic 3D tracking of every pixel in a video by utilizing a global 3D scene representation and novel 3D correlation scheme for dense flow est...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02573
• PDF: https://arxiv.org/pdf/2603.02573
• Project Page: https://jiah-cloud.github.io/Track4World.github.io/
• Github: https://github.com/TencentARC/Track4World

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
1
Beyond Language Modeling: An Exploration of Multimodal Pretraining

📝 Summary:
Controlled multimodal pretraining experiments reveal key insights about unified visual representations, data complementarity, world modeling emergence, and efficient scaling through mixture-of-experts...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03276
• PDF: https://arxiv.org/pdf/2603.03276

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
BeyondSWE: Can Current Code Agent Survive Beyond Single-Repo Bug Fixing?

📝 Summary:
Current code agent benchmarks fail to capture real-world complexity, prompting the creation of BeyondSWE to evaluate broader reasoning and knowledge scopes, alongside SearchSWE to study external knowl...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03194
• PDF: https://arxiv.org/pdf/2603.03194
• Project Page: https://aweai-team.github.io/BeyondSWE/
• Github: https://github.com/AweAI-Team/BeyondSWE

Datasets citing this paper:
https://huggingface.co/datasets/AweAI-Team/BeyondSWE

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
APRES: An Agentic Paper Revision and Evaluation System

📝 Summary:
Large language models are used to automatically revise scientific papers based on citation-predictive rubrics while preserving core content, achieving improved citation predictions and human evaluator...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03142
• PDF: https://arxiv.org/pdf/2603.03142

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Code2Math: Can Your Code Agent Effectively Evolve Math Problems Through Exploration?

📝 Summary:
Code agents can autonomously generate more complex mathematical problems by evolving existing ones, providing a scalable solution for creating high-difficulty reasoning problems. AI-generated summary ...

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03202
• PDF: https://arxiv.org/pdf/2603.03202
• Github: https://github.com/TarferSoul/Code2Math

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
1
Fast Matrix Multiplication in Small Formats: Discovering New Schemes with an Open-Source Flip Graph Framework

📝 Summary:
A new open-source C++ framework discovers fast matrix multiplication schemes, improving 79 ranks. It found a 4x4x10 scheme with 115 multiplications, beating Strassen's exponent for that size, and redistributes many schemes to simpler coefficients. Tools are public.

🔹 Publication Date: Published on Mar 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.02398
• PDF: https://arxiv.org/pdf/2603.02398
• Project Page: https://github.com/dronperminov/FastMatrixMultiplication
• Github: https://github.com/dronperminov/ternary_flip_graph

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
AgentConductor: Topology Evolution for Multi-Agent Competition-Level Code Generation

📝 Summary:
AgentConductor uses reinforcement learning-optimized multi-agent systems with an LLM-based orchestrator to dynamically generate interaction topologies for code generation, improving accuracy while red...

🔹 Publication Date: Published on Feb 19

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.17100
• PDF: https://arxiv.org/pdf/2602.17100

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
This media is not supported in your browser
VIEW IN TELEGRAM
Utonia: Toward One Encoder for All Point Clouds

📝 Summary:
Utonia introduces a unified self-supervised transformer encoder for diverse point cloud domains. It enhances perception and aids embodied and multimodal reasoning, aiming for foundation models in sparse 3D data.

🔹 Publication Date: Published on Mar 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.03283
• PDF: https://arxiv.org/pdf/2603.03283
• Project Page: https://pointcept.github.io/Utonia/
• Github: https://github.com/Pointcept/Utonia

🔹 Models citing this paper:
https://huggingface.co/Pointcept/Utonia

Spaces citing this paper:
https://huggingface.co/spaces/pointcept-bot/Utonia

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Qwen3-Coder-Next Technical Report

📝 Summary:
Qwen3-Coder-Next is an 80-billion-parameter language model that activates only 3 billion parameters during inference, achieving strong coding capabilities through agentic training with verifiable task...

🔹 Publication Date: Published on Feb 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00729
• PDF: https://arxiv.org/pdf/2603.00729
• Project Page: https://github.com/QwenLM/Qwen3-Coder
• Github: https://github.com/QwenLM/Qwen3-Coder

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning

📝 Summary:
AReaL, a fully asynchronous reinforcement learning system, decouples generation and training to achieve higher GPU utilization and up to 2.57x training speedup for large language models on reasoning t...

🔹 Publication Date: Published on May 30, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2505.24298
• PDF: https://arxiv.org/pdf/2505.24298
• Github: https://github.com/inclusionAI/AReaL

🔹 Models citing this paper:
https://huggingface.co/inclusionAI/AReaL-boba-2-8B
https://huggingface.co/inclusionAI/AReaL-boba-2-14B
https://huggingface.co/inclusionAI/AReaL-boba-2-8B-Open

Datasets citing this paper:
https://huggingface.co/datasets/inclusionAI/AReaL-tau2-data

Spaces citing this paper:
https://huggingface.co/spaces/rzvn/Medieval-Village-AI

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
InfoPO: Information-Driven Policy Optimization for User-Centric Agents

📝 Summary:
InfoPO optimizes agent-user collaboration for underspecified requests. It uses an information-gain reward to credit valuable turns that reduce uncertainty, improving decision-making and outperforming multi-turn RL baselines.

🔹 Publication Date: Published on Feb 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.00656
• PDF: https://arxiv.org/pdf/2603.00656
• Github: https://github.com/kfq20/InfoPO

==================================

For more data science resources:
https://t.iss.one/DataScienceT

#ReinforcementLearning #AI #HumanComputerInteraction #InformationTheory #AIagents