ML Research Hub
32.8K subscribers
4.34K photos
264 videos
23 files
4.69K links
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.

Admin: @HusseinSheikho || @Hussein_Sheikho
Download Telegram
Thinking with Map: Reinforced Parallel Map-Augmented Agent for Geolocalization

📝 Summary:
Large vision-language models are enhanced for image geolocalization by incorporating map-based reasoning and agent-in-the-map loop optimization, achieving superior accuracy compared to existing models...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05432
• PDF: https://arxiv.org/pdf/2601.05432
• Project Page: https://amap-ml.github.io/Thinking-with-Map/
• Github: https://github.com/AMAP-ML/Thinking-with-Map

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
The Molecular Structure of Thought: Mapping the Topology of Long Chain-of-Thought Reasoning

📝 Summary:
Large language models struggle with long chain-of-thought reasoning due to unstable structural patterns, but a molecular-inspired approach using effective semantic isomers and distribution-transfer-gr...

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.06002
• PDF: https://arxiv.org/pdf/2601.06002

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Can We Predict Before Executing Machine Learning Agents?

📝 Summary:
Autonomous machine learning agents overcome execution bottlenecks by predicting outcomes before physical execution, achieving faster convergence and improved performance through a predict-then-verify ...

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05930
• PDF: https://arxiv.org/pdf/2601.05930
• Github: https://github.com/zjunlp/predict-before-execute

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency

📝 Summary:
Large language models exhibit brittle beliefs under contextual perturbations, which are better measured by structural consistency metrics and addressed through structure-aware training methods. AI-gen...

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05905
• PDF: https://arxiv.org/pdf/2601.05905
• Github: https://github.com/zjunlp/belief

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Orient Anything V2: Unifying Orientation and Rotation Understanding

📝 Summary:
Orient Anything V2 enhances 3D orientation understanding through scalable 3D asset synthesis, symmetry-aware periodic distribution fitting, and multi-frame relative rotation prediction, achieving stat...

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05573
• PDF: https://arxiv.org/pdf/2601.05573
• Project Page: https://orient-anythingv2.github.io/
• Github: https://github.com/SpatialVision/Orient-Anything-V2

🔹 Models citing this paper:
https://huggingface.co/Viglong/OriAnyV2_ckpt

Datasets citing this paper:
https://huggingface.co/datasets/Viglong/OriAnyV2_Train_Render

Spaces citing this paper:
https://huggingface.co/spaces/Viglong/Orient-Anything-V2

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
SmartSearch: Process Reward-Guided Query Refinement for Search Agents

📝 Summary:
SmartSearch enhances LLM-based search agents through process rewards and query refinement mechanisms that improve intermediate search query quality via a three-stage curriculum learning approach. AI-g...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04888
• PDF: https://arxiv.org/pdf/2601.04888
• Github: https://github.com/MYVAE/SmartSearch?tab=readme-ov-file

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Router-Suggest: Dynamic Routing for Multimodal Auto-Completion in Visually-Grounded Dialogs

📝 Summary:
Multimodal auto-completion leverages visual and textual context to improve real-time prediction accuracy in conversational interfaces, with a router framework enabling efficient model selection based ...

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05851
• PDF: https://arxiv.org/pdf/2601.05851

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
AgentOCR: Reimagining Agent History via Optical Self-Compression

📝 Summary:
AgentOCR reimagines agent history as visual tokens to reduce token consumption and memory in agentic systems. It leverages optical caching and adaptive self-compression. This framework maintains strong performance while significantly cutting token usage and boosting efficiency.

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04786
• PDF: https://arxiv.org/pdf/2601.04786

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
MMFormalizer: Multimodal Autoformalization in the Wild

📝 Summary:
MMFormalizer enables multimodal autoformalization by integrating visual perception with formal mathematical reasoning, supporting complex physical domains from classical mechanics to quantum mechanics...

🔹 Publication Date: Published on Jan 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03017
• PDF: https://arxiv.org/pdf/2601.03017
• Project Page: https://mmformalizer.github.io/

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Qwen3-VL-Embedding and Qwen3-VL-Reranker: A Unified Framework for State-of-the-Art Multimodal Retrieval and Ranking

📝 Summary:
The Qwen3-VL-Embedding and Qwen3-VL-Reranker models form an end-to-end multimodal search pipeline, leveraging multi-stage training and cross-attention mechanisms to achieve high-precision retrieval ac...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://www.arxiv.org/abs/2601.04720
• PDF: https://arxiv.org/pdf/2601.04720
• Github: https://github.com/QwenLM/Qwen3-VL-Embedding

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
AnyDepth: Depth Estimation Made Easy

📝 Summary:
A lightweight monocular depth estimation framework uses DINOv3 as visual encoder and a compact transformer decoder to achieve higher accuracy with reduced computational overhead and improved data qual...

🔹 Publication Date: Published on Jan 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02760
• PDF: https://arxiv.org/pdf/2601.02760
• Project Page: https://aigeeksgroup.github.io/AnyDepth
• Github: https://aigeeksgroup.github.io/AnyDepth

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
CaricatureGS: Exaggerating 3D Gaussian Splatting Faces With Gaussian Curvature

📝 Summary:
CaricatureGS introduces a 3D caricaturization framework combining Gaussian curvature-based exaggeration with 3D Gaussian Splatting for photorealistic, controllable face avatars. It uses a unique training scheme with synthesized supervision to achieve high fidelity, real-time deformation, and cont...

🔹 Publication Date: Published on Jan 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03319
• PDF: https://arxiv.org/pdf/2601.03319
• Project Page: https://c4ricaturegs.github.io/
• Github: https://c4ricaturegs.github.io/

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
Memory Matters More: Event-Centric Memory as a Logic Map for Agent Searching and Reasoning

📝 Summary:
CompassMem is an event-centric memory framework that organizes experiences into an Event Graph to enable structured memory navigation and long-horizon reasoning beyond traditional retrieval methods. A...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04726
• PDF: https://arxiv.org/pdf/2601.04726

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research
FAPO: Flawed-Aware Policy Optimization for Efficient and Reliable Reasoning

📝 Summary:
FAPO improves reinforcement learning for LLMs by penalizing flawed-positive rollouts that reinforce unreliable reasoning. It uses these flaws for initial gains while shifting optimization toward reliable reasoning, enhancing correctness and stability.

🔹 Publication Date: Published on Oct 26, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2510.22543
• PDF: https://arxiv.org/pdf/2510.22543
• Project Page: https://fapo-rl.github.io/
• Github: https://fapo-rl.github.io

🔹 Models citing this paper:
https://huggingface.co/dyyyyyyyy/FAPO-GenRM-4B
https://huggingface.co/dyyyyyyyy/FAPO-32B

Datasets citing this paper:
https://huggingface.co/datasets/dyyyyyyyy/FAPO-Reasoning-Dataset
https://huggingface.co/datasets/dyyyyyyyy/FAPO-Critic

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

For more data science resources:
https://t.me/DataScienceT

#ReinforcementLearning #LLMs #AI #MachineLearning #Reasoning
Distilling Feedback into Memory-as-a-Tool

📝 Summary:
This framework converts transient critiques into retrievable guidelines using a file-based memory system and agent tools. It enables LLMs to achieve test-time refinement performance with significantly reduced inference costs.

🔹 Publication Date: Published on Jan 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05960
• PDF: https://arxiv.org/pdf/2601.05960
• Github: https://github.com/vicgalle/feedback-memory-as-a-tool

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

For more data science resources:
https://t.me/DataScienceT

#LLMs #AIAgents #MemorySystems #AIResearch #MachineLearning
Same Claim, Different Judgment: Benchmarking Scenario-Induced Bias in Multilingual Financial Misinformation Detection

📝 Summary:
A new benchmark, mfmdscen, evaluates behavioral biases in large language models for multilingual financial misinformation detection. It uses complex economic scenarios and a multilingual dataset, revealing significant biases across 22 mainstream LLMs.

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05403
• PDF: https://arxiv.org/pdf/2601.05403

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

For more data science resources:
https://t.me/DataScienceT

#LLM #AIbias #FinancialAI #MisinformationDetection #MultilingualAI
OmniFlatten: An End-to-end GPT Model for Seamless Voice Conversation

📝 Summary:
OmniFlatten is a GPT-based model for real-time, natural full-duplex spoken dialogue. It uses a multi-stage post-training method to adapt a text LLM for speech and text generation without altering its architecture, enabling low-latency conversations.

🔹 Publication Date: Published on Oct 23, 2024

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2410.17799
• PDF: https://arxiv.org/pdf/2410.17799
• Github: https://github.com/karpathy/nanogpt

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

For more data science resources:
https://t.me/DataScienceT

#GPT #VoiceAI #LLM #RealTimeAI #NLP
TCAndon-Router: Adaptive Reasoning Router for Multi-Agent Collaboration

📝 Summary:
TCAndon-Router TCAR is an adaptive reasoning router for multi-agent systems. It overcomes limitations of existing task routers by supporting dynamic agent onboarding and generating natural language reasoning chains to select agents. TCAR significantly improves routing accuracy, reduces conflicts,...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04544
• PDF: https://arxiv.org/pdf/2601.04544
• Github: https://github.com/Tencent/TCAndon-Router

🔹 Models citing this paper:
https://huggingface.co/tencent/TCAndon-Router

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

For more data science resources:
https://t.me/DataScienceT

#MultiAgentSystems #AI #NLP #AdaptiveSystems #AIResearch
Media is too big
VIEW IN TELEGRAM
NitroGen: An Open Foundation Model for Generalist Gaming Agents

📝 Summary:
NitroGen is a vision-action foundation model trained on extensive gameplay data that demonstrates strong cross-game generalization and effective transfer learning capabilities. AI-generated summary We...

🔹 Publication Date: Published on Jan 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02427
• PDF: https://arxiv.org/pdf/2601.02427
• Project Page: https://nitrogen.minedojo.org/
• Github: https://github.com/MineDojo/NitroGen

🔹 Models citing this paper:
https://huggingface.co/nvidia/NitroGen

Datasets citing this paper:
https://huggingface.co/datasets/nvidia/NitroGen

Spaces citing this paper:
https://huggingface.co/spaces/dennny123/NitroGen-SuperstarSaga
https://huggingface.co/spaces/blanchon/NitroGen-Pokemon

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

For more data science resources:
https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research