ML Research Hub
32.9K subscribers
5.35K photos
332 videos
24 files
5.78K links
Advancing research in Machine Learning – practical insights, tools, and techniques for researchers.

Admin: @HusseinSheikho || @Hussein_Sheikho
Download Telegram
Latent Thoughts Tuning: Bridging Context and Reasoning with Fused Information in Latent Tokens

📝 Summary:
Latent Thoughts Tuning introduces a novel framework for robust reasoning in continuous latent space. It addresses feature collapse by fusing contextual hidden states with predictive semantic guidance. This method outperforms baselines and achieves improved reasoning accuracy.

🔹 Publication Date: Published on Feb 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.10229
• PDF: https://arxiv.org/pdf/2602.10229
• Github: https://github.com/NeosKnight233/Latent-Thoughts-Tuning

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
LiveMedBench: A Contamination-Free Medical Benchmark for LLMs with Automated Rubric Evaluation

📝 Summary:
LiveMedBench addresses limitations in medical LLM evaluation by providing a continuously updated, contamination-free benchmark with rubric-based evaluation that better aligns with expert clinical reas...

🔹 Publication Date: Published on Feb 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.10367
• PDF: https://arxiv.org/pdf/2602.10367
• Project Page: https://zhilingyan.github.io/LiveMedBench/
• Github: https://github.com/ZhilingYan/LiveMedBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
TIC-VLA: A Think-in-Control Vision-Language-Action Model for Robot Navigation in Dynamic Environments

📝 Summary:
TIC-VLA is a latency-aware framework enhancing robot navigation by explicitly modeling delayed semantic reasoning. It uses a delayed semantic-control interface and latency-consistent training. This allows robots to maintain real-time control despite significant reasoning delays.

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02459
• PDF: https://arxiv.org/pdf/2602.02459
• Project Page: https://ucla-mobility.github.io/TIC-VLA/
• Github: https://github.com/ucla-mobility/TIC-VLA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
From Features to Actions: Explainability in Traditional and Agentic AI Systems

📝 Summary:
This paper compares static and agentic AI explainability. It finds attribution methods reliable for static predictions but not for diagnosing failures in multi-step agentic systems. Trace-based diagnostics effectively localize agentic breakdowns, urging a shift to trajectory-level explainability.

🔹 Publication Date: Published on Feb 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.06841
• PDF: https://arxiv.org/pdf/2602.06841
• Project Page: https://vectorinstitute.github.io/unified-xai-evaluation-framework/
• Github: https://github.com/VectorInstitute/unified-xai-evaluation-framework

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Media is too big
VIEW IN TELEGRAM
GigaBrain-0.5M*: a VLA That Learns From World Model-Based Reinforcement Learning

📝 Summary:
GigaBrain-0.5M enhances vision-language-action models by integrating world model-based reinforcement learning. This improves performance by 30% on complex robotic tasks and enables reliable long-horizon execution, overcoming prior VLA limitations.

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.12099
• PDF: https://arxiv.org/pdf/2602.12099
• Project Page: https://gigabrain05m.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolation

📝 Summary:
On-policy distillation is extended through a generalized framework that introduces flexible reference models and reward scaling factors, demonstrating improved performance through reward extrapolation...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.12125
• PDF: https://arxiv.org/pdf/2602.12125
• Github: https://github.com/RUCBM/G-OPD

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Unveiling Implicit Advantage Symmetry: Why GRPO Struggles with Exploration and Difficulty Adaptation

📝 Summary:
Asymmetric Group Relative Advantage Estimation addresses exploration and difficulty adaptation challenges in reinforcement learning with large language models by dynamically modulating exploration inc...

🔹 Publication Date: Published on Feb 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.05548
• PDF: https://arxiv.org/pdf/2602.05548
• Github: https://github.com/HKU-HealthAI/A-GRAE

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Budget-Constrained Agentic Large Language Models: Intention-Based Planning for Costly Tool Use

📝 Summary:
Budget-constrained tool-augmented agents use a hierarchical world model and intent-aware planning to optimize multi-step task completion under monetary constraints. AI-generated summary We study budge...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11541
• PDF: https://arxiv.org/pdf/2602.11541

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Dreaming in Code for Curriculum Learning in Open-Ended Worlds

📝 Summary:
Foundation models generate executable environment code to scaffold learning progress in open-ended worlds, enabling agents to acquire long-horizon skills through curriculum control. AI-generated summa...

🔹 Publication Date: Published on Feb 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.08194
• PDF: https://arxiv.org/pdf/2602.08194
• Project Page: https://konstantinosmitsides.github.io/dreaming-in-code
• Github: https://github.com/konstantinosmitsides/dreaming-in-code

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Neural Additive Experts: Context-Gated Experts for Controllable Model Additivity

📝 Summary:
Neural Additive Experts combines multiple specialized networks with a dynamic gating mechanism to balance predictive accuracy and feature interpretability in machine learning models. AI-generated summ...

🔹 Publication Date: Published on Feb 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.10585
• PDF: https://arxiv.org/pdf/2602.10585
• Github: https://github.com/Teddy-XiongGZ/NAE

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
The Devil Behind Moltbook: Anthropic Safety is Always Vanishing in Self-Evolving AI Societies

📝 Summary:
Multi-agent LLM systems cannot achieve continuous self-improvement and maintain safety if isolated. Isolated self-evolution causes statistical blind spots, leading to irreversible safety degradation. This is a fundamental limit, requiring external oversight or new safety mechanisms.

🔹 Publication Date: Published on Feb 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.09877
• PDF: https://arxiv.org/pdf/2602.09877

Datasets citing this paper:
https://huggingface.co/datasets/xunyoyo/Self-Evolving-Safety

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
LawThinker: A Deep Research Legal Agent in Dynamic Environments

📝 Summary:
LawThinker is an autonomous legal research agent that uses an Explore-Verify-Memorize strategy with a DeepVerifier module to ensure accurate and procedurally compliant legal reasoning through dynamic ...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.12056
• PDF: https://arxiv.org/pdf/2602.12056
• Github: https://github.com/yxy-919/LawThinker-agent

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Voxtral Realtime

📝 Summary:
Voxtral Realtime is a streaming speech recognition model trained end-to-end for sub-second latency with performance matching offline systems. AI-generated summary We introduce Voxtral Realtime, a nati...

🔹 Publication Date: Published on Feb 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11298
• PDF: https://arxiv.org/pdf/2602.11298
• Project Page: https://mistral.ai/news/voxtral-transcribe-2

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
dVoting: Fast Voting for dLLMs

📝 Summary:
Diffusion large language models enable parallel token generation and efficient reasoning enhancement through a voting technique that identifies and refines uncertain predictions across multiple sample...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.12153
• PDF: https://arxiv.org/pdf/2602.12153
• Github: https://github.com/fscdc/dVoting

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces

📝 Summary:
ThinkRouter is a confidence-aware routing mechanism that improves reasoning efficiency by switching between discrete token and latent spaces based on model confidence, achieving better accuracy and fa...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11683
• PDF: https://arxiv.org/pdf/2602.11683

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
ScalSelect: Scalable Training-Free Multimodal Data Selection for Efficient Visual Instruction Tuning

📝 Summary:
ScalSelect is a scalable training-free method for selecting representative multimodal data that achieves near-full-dataset performance with significantly reduced computational requirements. AI-generat...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11636
• PDF: https://arxiv.org/pdf/2602.11636
• Github: https://github.com/ChangtiWu/ScalSelect

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

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
MolmoSpaces: A Large-Scale Open Ecosystem for Robot Navigation and Manipulation

📝 Summary:
MolmoSpaces presents an open ecosystem with diverse indoor environments and annotated objects for large-scale robot policy benchmarking across multiple tasks and simulators. AI-generated summary Deplo...

🔹 Publication Date: Published on Feb 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11337
• PDF: https://arxiv.org/pdf/2602.11337
• Project Page: https://allenai.org/blog/molmospaces
• Github: https://github.com/allenai/molmospaces

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments

📝 Summary:
Gaia2 presents a benchmark for evaluating large language model agents in asynchronous, dynamic environments with temporal constraints and multi-agent collaboration, featuring a write-action verifier f...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11964
• PDF: https://arxiv.org/pdf/2602.11964

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research
MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling

📝 Summary:
MiniCPM-SALA combines sparse and linear attention mechanisms in a hybrid architecture to enable efficient processing of ultra-long contexts while maintaining model performance and reducing training co...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11761
• PDF: https://arxiv.org/pdf/2602.11761
• Github: https://github.com/OpenBMB/MiniCPM

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

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
ABot-N0: Technical Report on the VLA Foundation Model for Versatile Embodied Navigation

📝 Summary:
A unified Vision-Language-Action model with a hierarchical architecture combining semantic reasoning and continuous trajectory generation achieves state-of-the-art performance across multiple embodied...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.11598
• PDF: https://arxiv.org/pdf/2602.11598
• Project Page: https://amap-cvlab.github.io/ABot-Navigation/ABot-N0/
• Github: https://github.com/amap-cvlab/ABot-Navigation/tree/ABot-N0

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

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
Stroke of Surprise: Progressive Semantic Illusions in Vector Sketching

📝 Summary:
Progressive Semantic Illusions use a generative framework with dual-branch Score Distillation Sampling to create vector sketches that transform semantically through sequential stroke additions, achiev...

🔹 Publication Date: Published on Feb 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.12280
• PDF: https://arxiv.org/pdf/2602.12280
• Project Page: https://stroke-of-surprise.github.io/
• Github: https://github.com/stroke-of-surprise/Stroke-Of-Surprise

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research