✨Sparse Reward Subsystem in Large Language Models

📝 Summary:
Research identifies a sparse reward subsystem in LLM hidden states containing value neurons that represent internal state expectations and dopamine-like neurons encoding reward prediction errors. AI-g...

🔹 Publication Date: Published on Feb 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.00986
• PDF: https://arxiv.org/pdf/2602.00986

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Hunt Instead of Wait: Evaluating Deep Data Research on Large Language Models

📝 Summary:
Agentic large language models require investigatory intelligence for autonomous data analysis, demonstrated through the Deep Data Research benchmark that evaluates their ability to extract insights fr...

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02039
• PDF: https://arxiv.org/pdf/2602.02039
• Project Page: https://huggingface.co/spaces/thinkwee/DDR_Bench
• Github: https://github.com/thinkwee/DDR_Bench

✨ Datasets citing this paper:
• https://huggingface.co/datasets/thinkwee/DDRBench_10K

✨ Spaces citing this paper:
• https://huggingface.co/spaces/thinkwee/DDR_Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Rethinking Generative Recommender Tokenizer: Recsys-Native Encoding and Semantic Quantization Beyond LLMs

📝 Summary:
ReSID introduces a recommendation-native framework to improve sequential recommenders. It learns predictive item representations and optimizes quantization for better information preservation and sequential predictability without LLMs. ReSID significantly outperforms baselines by over 10% and red...

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02338
• PDF: https://arxiv.org/pdf/2602.02338

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Clipping-Free Policy Optimization for Large Language Models

📝 Summary:
Clipping-Free Policy Optimization replaces heuristic clipping with convex quadratic penalty to stabilize reinforcement learning training for large language models without performance loss. AI-generate...

🔹 Publication Date: Published on Jan 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22801
• PDF: https://arxiv.org/pdf/2601.22801

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Gaming the Judge: Unfaithful Chain-of-Thought Can Undermine Agent Evaluation

📝 Summary:
Large language models used as judges for agent performance evaluation are vulnerable to manipulation of reasoning traces, with content-based fabrications being more effective than style-based alterati...

🔹 Publication Date: Published on Jan 21

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.14691
• PDF: https://arxiv.org/pdf/2601.14691

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation

📝 Summary:
Automated pipeline for sound separation using high-purity single-event segments from in-the-wild datasets achieves competitive performance with significantly reduced data requirements. AI-generated su...

🔹 Publication Date: Published on Jan 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22599
• PDF: https://arxiv.org/pdf/2601.22599
• Project Page: https://shandaai.github.io/Hive
• Github: https://github.com/ShandaAI/Hive

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Where to Attend: A Principled Vision-Centric Position Encoding with Parabolas

📝 Summary:
Parabolic Position Encoding (PaPE) is a novel position encoding method for vision modalities that improves upon existing approaches by incorporating translation invariance, rotation invariance, distan...

🔹 Publication Date: Published on Feb 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.01418
• PDF: https://arxiv.org/pdf/2602.01418
• Project Page: https://chrisohrstrom.github.io/parabolic-position-encoding/
• Github: https://github.com/DTU-PAS/parabolic-position-encoding

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨YOLOE-26: Integrating YOLO26 with YOLOE for Real-Time Open-Vocabulary Instance Segmentation

📝 Summary:
YOLOE-26 integrates YOLO26 architecture with open-vocabulary learning for real-time instance segmentation, utilizing convolutional backbones, end-to-end regression, and object embedding heads with tex...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.00168
• PDF: https://arxiv.org/pdf/2602.00168

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Training LLMs for Divide-and-Conquer Reasoning Elevates Test-Time Scalability

📝 Summary:
A new reinforcement learning framework trains LLMs for divide-and-conquer reasoning. This method decomposes complex problems, significantly elevating test-time scalability and outperforming traditional chain-of-thought approaches on challenging benchmarks.

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02477
• PDF: https://arxiv.org/pdf/2602.02477
• Github: https://github.com/MasterVito/DAC-RL

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨CUA-Skill: Develop Skills for Computer Using Agent

📝 Summary:
CUA-Skill introduces a large-scale library of engineered computer-use skills that enhance agent performance and efficiency on Windows-based tasks. AI-generated summary Computer-Using Agents (CUAs) aim...

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21123
• PDF: https://arxiv.org/pdf/2601.21123
• Project Page: https://microsoft.github.io/cua_skill/
• Github: https://github.com/microsoft/cua_skill

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨VoxServe: Streaming-Centric Serving System for Speech Language Models

📝 Summary:
VoxServe is a unified serving system for SpeechLMs that optimizes streaming performance. It uses model-execution abstraction, streaming-aware scheduling, and asynchronous inference pipelines. This achieves 10-20x higher throughput at comparable latency for diverse SpeechLM architectures.

🔹 Publication Date: Published on Jan 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.00269
• PDF: https://arxiv.org/pdf/2602.00269
• Project Page: https://vox-serve.github.io/
• Github: https://github.com/vox-serve/vox-serve

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Competing Visions of Ethical AI: A Case Study of OpenAI

📝 Summary:
This study analyzed OpenAI's public discourse on ethical AI. It found OpenAI primarily frames the discussion around safety and risk, largely avoiding academic ethics frameworks. This indicates a distinct approach to AI ethics in industry communications.

🔹 Publication Date: Published on Jan 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.16513
• PDF: https://arxiv.org/pdf/2601.16513

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨CodeOCR: On the Effectiveness of Vision Language Models in Code Understanding

📝 Summary:
Multimodal LLMs can effectively understand source code represented as compressed images, achieving up to 8x token reduction. This method leverages visual cues and sometimes outperforms text inputs, indicating a path to more efficient code comprehension.

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.01785
• PDF: https://arxiv.org/pdf/2602.01785

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Research on World Models Is Not Merely Injecting World Knowledge into Specific Tasks

📝 Summary:
Current world models lack unified frameworks despite task-specific advances, necessitating a comprehensive approach integrating interaction, perception, symbolic reasoning, and spatial representation....

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.01630
• PDF: https://arxiv.org/pdf/2602.01630
• Github: https://github.com/OpenDCAI/DataFlow-MM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AdaptMMBench: Benchmarking Adaptive Multimodal Reasoning for Mode Selection and Reasoning Process

📝 Summary:
AdaptMMBench presents a comprehensive benchmark for evaluating adaptive multimodal reasoning in Vision-Language Models, measuring reasoning mode selection rationality through dynamic difficulty assess...

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02676
• PDF: https://arxiv.org/pdf/2602.02676

✨ Datasets citing this paper:
• https://huggingface.co/datasets/xintongzhang/AdaptMMBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Unified Personalized Reward Model for Vision Generation

📝 Summary:
UnifiedReward-Flex combines reward modeling with flexible, context-adaptive reasoning to improve visual generation by dynamically constructing hierarchical assessments based on semantic intent and vis...

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02380
• PDF: https://arxiv.org/pdf/2602.02380
• Project Page: https://codegoat24.github.io/UnifiedReward/flex
• Github: https://codegoat24.github.io/UnifiedReward/flex

🔹 Models citing this paper:
• https://huggingface.co/CodeGoat24/Wan2.1-T2V-14B-UnifiedReward-Flex-lora
• https://huggingface.co/CodeGoat24/UnifiedReward-Flex-qwen3vl-2b
• https://huggingface.co/CodeGoat24/UnifiedReward-Flex-qwen3vl-4b

✨ Datasets citing this paper:
• https://huggingface.co/datasets/CodeGoat24/UnifiedReward-Flex-SFT-90K

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Glance and Focus Reinforcement for Pan-cancer Screening

📝 Summary:
A reinforcement learning framework with glance and focus models improves pan-cancer screening in CT scans by addressing foreground-background imbalance and reducing false positives through group relat...

🔹 Publication Date: Published on Jan 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.19103
• PDF: https://arxiv.org/pdf/2601.19103
• Github: https://github.com/Luffy03/GF-Screen

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨FaceLinkGen: Rethinking Identity Leakage in Privacy-Preserving Face Recognition with Identity Extraction

📝 Summary:
FaceLinkGen attack demonstrates that current privacy-preserving face recognition methods fail to protect identity information despite pixel-level distortion metrics suggesting adequate protection. AI-...

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.02914
• PDF: https://arxiv.org/pdf/2602.02914

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ObjEmbed: Towards Universal Multimodal Object Embeddings

📝 Summary:
ObjEmbed is a novel multimodal language-model embedding approach that decomposes images into regional embeddings for improved object-level visual understanding and retrieval tasks. AI-generated summar...

🔹 Publication Date: Published on Feb 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.01753
• PDF: https://arxiv.org/pdf/2602.01753
• Github: https://github.com/WeChatCV/ObjEmbed

🔹 Models citing this paper:
• https://huggingface.co/fushh7/ObjEmbed-2B
• https://huggingface.co/fushh7/ObjEmbed-4B

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learning Query-Specific Rubrics from Human Preferences for DeepResearch Report Generation

📝 Summary:
DeepResearch report generation is improved via human-preference-aligned, query-specific rubric generators trained with reinforcement learning and a multi-agent workflow. This system significantly outperforms open-source baselines and matches leading closed-source models.

🔹 Publication Date: Published on Feb 3

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.03619
• PDF: https://arxiv.org/pdf/2602.03619

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research