✨What Matters in Data Curation for Multimodal Reasoning? Insights from the DCVLR Challenge

📝 Summary:
Data curation for multimodal reasoning shows that difficulty-based example selection on aligned datasets drives performance gains, while increasing dataset size mainly reduces variance and synthetic a...

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.10922
• PDF: https://arxiv.org/pdf/2601.10922

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge

📝 Summary:
Multiplex Thinking is a stochastic soft reasoning method that samples and aggregates multiple candidate tokens at each step into a single multiplex token. This allows it to compactly represent multiple plausible next steps, resulting in shorter sequences than Chain-of-Thought. It outperforms CoT ...

🔹 Publication Date: Published on Jan 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.08808
• PDF: https://arxiv.org/pdf/2601.08808
• Project Page: https://gmlr-penn.github.io/Multiplex-Thinking/
• Github: https://github.com/GMLR-Penn/Multiplex-Thinking

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ABC-Bench: Benchmarking Agentic Backend Coding in Real-World Development

📝 Summary:
ABC-Bench evaluates LLM agents on realistic backend coding tasks requiring full development lifecycle management from repository exploration to containerized service deployment and API testing. AI-gen...

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11077
• PDF: https://arxiv.org/pdf/2601.11077

🔹 Models citing this paper:
• https://huggingface.co/OpenMOSS-Team/Qwen3-32B-ABC
• https://huggingface.co/OpenMOSS-Team/Qwen3-8B-ABC

✨ Datasets citing this paper:
• https://huggingface.co/datasets/OpenMOSS-Team/ABC-Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models

📝 Summary:
Research reveals that large language models operate within a persona space where an "Assistant Axis" controls helpfulness and behavioral stability, with steering techniques able to influence model res...

🔹 Publication Date: Published on Jan 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.10387
• PDF: https://arxiv.org/pdf/2601.10387

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨CoDance: An Unbind-Rebind Paradigm for Robust Multi-Subject Animation

📝 Summary:
CoDance introduces an Unbind-Rebind framework for animating multiple subjects with flexible spatial configurations, using pose shift encoding and semantic/textual guidance for motion reassignment. AI-...

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11096
• PDF: https://arxiv.org/pdf/2601.11096
• Project Page: https://lucaria-academy.github.io/CoDance/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EverMemOS: A Self-Organizing Memory Operating System for Structured Long-Horizon Reasoning

📝 Summary:
EverMemOS is a self-organizing memory system for LLMs that processes dialogue into structured memory cells and scenes. This enhances long-term interaction and reasoning, achieving state-of-the-art performance.

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02163
• PDF: https://arxiv.org/pdf/2601.02163
• Github: https://github.com/EverMind-AI/EverMemOS

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Continuous Audio Language Models

📝 Summary:
Audio Language Models (ALM) have emerged as the dominant paradigm for speech and music generation by representing audio as sequences of discrete tokens. Yet, unlike text tokens, which are invertible, ...

🔹 Publication Date: Published on Sep 8, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2509.06926
• PDF: https://arxiv.org/pdf/2509.06926
• Project Page: https://huggingface.co/spaces/kyutai/calm-samples
• Github: https://github.com/kyutai-labs/pocket-tts

🔹 Models citing this paper:
• https://huggingface.co/kyutai/pocket-tts
• https://huggingface.co/kyutai/pocket-tts-without-voice-cloning
• https://huggingface.co/Verylicious/pocket-tts-ungated

✨ Spaces citing this paper:
• https://huggingface.co/spaces/D3vShoaib/pocket-tts
• https://huggingface.co/spaces/Xlnk/tts
• https://huggingface.co/spaces/kyutai/calm-samples

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨InfiAgent: An Infinite-Horizon Framework for General-Purpose Autonomous Agents

📝 Summary:
InfiAgent is a framework that maintains bounded reasoning context for long-horizon tasks by externalizing persistent state into a file-centric abstraction, enabling stable performance without task-spe...

🔹 Publication Date: Published on Jan 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03204
• PDF: https://arxiv.org/pdf/2601.03204
• Github: https://github.com/ChenglinPoly/infiAgent

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MemOS: A Memory OS for AI System

📝 Summary:
MemOS is a memory operating system for LLMs that unifies plaintext, activation-based, and parameter-level memories. It treats memory as a system resource, using MemCubes for efficient storage, retrieval, and enabling continual learning and personalized modeling.

🔹 Publication Date: Published on Jul 4, 2025

🔹 Paper Links:
• arXiv Page: https://arxivlens.com/PaperView/Details/memos-a-memory-os-for-ai-system-4846-c5e0c676
• PDF: https://arxiv.org/pdf/2507.03724
• Project Page: https://memos.openmem.net/
• Github: https://github.com/MemTensor/MemOS

🔹 Models citing this paper:
• https://huggingface.co/kagvi13/HMP

✨ Datasets citing this paper:
• https://huggingface.co/datasets/MemTensor/MemOS_eval_result

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

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

#AI #LLMs #MemoryOS #ContinualLearning #SystemDesign

✨SIN-Bench: Tracing Native Evidence Chains in Long-Context Multimodal Scientific Interleaved Literature

📝 Summary:
SIN-Bench and the Fish-in-the-Ocean paradigm evaluate multimodal models by requiring explicit cross-modal evidence chains in scientific documents. This new method shows grounding is a bottleneck, revealing a gap between answer correctness and traceable support.

🔹 Publication Date: Published on Jan 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.10108
• PDF: https://arxiv.org/pdf/2601.10108

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

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

#MultimodalAI #Benchmarking #ScientificLiterature #AIResearch #Grounding

✨Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMs

📝 Summary:
Spurious rewards in RLVR trigger a memorization shortcut in LLMs, causing a paradox where models bypass reasoning. Researchers identified a hidden neural Anchor-Adapter circuit enabling this shortcut. This discovery offers a way to mitigate data contamination in RLVR models.

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11061
• PDF: https://arxiv.org/pdf/2601.11061
• Github: https://github.com/idwts/How-RLVR-Activates-Memorization-Shortcuts

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

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

#LLMs #ReinforcementLearning #SpuriousRewards #AIResearch #DataContamination

✨YaPO: Learnable Sparse Activation Steering Vectors for Domain Adaptation

📝 Summary:
YaPO learns sparse steering vectors for LLMs using Sparse Autoencoders, enabling more effective and stable control than dense methods. This leads to disentangled, interpretable directions for fine-grained alignment across various behaviors, without degrading general knowledge. YaPO offers a gener...

🔹 Publication Date: Published on Jan 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.08441
• PDF: https://arxiv.org/pdf/2601.08441
• Project Page: https://mbzuai-paris.github.io/YaPO/
• Github: https://github.com/MBZUAI-Paris/YaPO

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

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

#LLMs #DomainAdaptation #SparseLearning #MachineLearning #AI

✨Medical SAM3: A Foundation Model for Universal Prompt-Driven Medical Image Segmentation

📝 Summary:
Medical SAM3 is a foundation model for universal prompt-driven medical image segmentation. It fine-tunes the general SAM3 on diverse medical datasets to overcome domain shifts. This provides robust, flexible segmentation across modalities and structures.

🔹 Publication Date: Published on Jan 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.10880
• PDF: https://arxiv.org/pdf/2601.10880
• Project Page: https://chongcongjiang.github.io/MedicalSAM3/
• Github: https://github.com/AIM-Research-Lab/Medical-SAM3.git

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

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

#MedicalAI #ImageSegmentation #FoundationModels #DeepLearning #ComputerVision

✨CLARE: Continual Learning for Vision-Language-Action Models via Autonomous Adapter Routing and Expansion

📝 Summary:
CLARE enables robots to continually learn new tasks without forgetting, using lightweight adapters. It autonomously expands these adapters and dynamically routes them, ensuring high performance without needing task labels or storing past data.

🔹 Publication Date: Published on Jan 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.09512
• PDF: https://arxiv.org/pdf/2601.09512
• Project Page: https://tum-lsy.github.io/clare/
• Github: https://github.com/utiasDSL/clare

✨ Datasets citing this paper:
• https://huggingface.co/datasets/continuallearning/libero_10_image_task_0

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

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

#ContinualLearning #Robotics #AI #MachineLearning #VLAModels

✨PubMed-OCR: PMC Open Access OCR Annotations

📝 Summary:
PubMed-OCR is a corpus of 209.5K scientific articles from PubMed Central with Google Cloud Vision OCR annotations. It provides word, line, and paragraph bounding boxes to support layout-aware modeling and OCR evaluation. This data is publicly released.

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11425
• PDF: https://arxiv.org/pdf/2601.11425

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

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

#OCR #Dataset #ComputerVision #MachineLearning #DataScience

✨NAACL: Noise-AwAre Verbal Confidence Calibration for LLMs in RAG Systems

📝 Summary:
LLMs in RAG systems exhibit poor confidence calibration due to noisy contexts. This paper proposes NAACL, a noise-aware calibration framework. NAACL uses new rules and supervised fine-tuning to make LLMs intrinsically aware of noisy input, significantly improving confidence calibration.

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11004
• PDF: https://arxiv.org/pdf/2601.11004
• Github: https://github.com/HKUST-KnowComp/NAACL

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

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

#LLMs #RAG #ConfidenceCalibration #NLP #AI

✨PaddleOCR 3.0 Technical Report

📝 Summary:
PaddleOCR 3.0 is an open-source toolkit offering efficient OCR and document parsing solutions. Its models achieve competitive accuracy and efficiency with fewer than 100 million parameters, rivaling much larger vision-language models.

🔹 Publication Date: Published on Jul 8, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2507.05595
• PDF: https://huggingface.co/collections/PaddlePaddle/pp-structurev3

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Multi-Agent Collaboration via Evolving Orchestration

📝 Summary:
A centralized orchestrator, trained with reinforcement learning, dynamically directs LLM agents for multi-agent collaboration. This puppeteer-style method achieves superior performance and reduced computational costs.

🔹 Publication Date: Published on May 26, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2505.19591
• PDF: https://arxiv.org/pdf/2505.19591
• Github: https://github.com/OpenBMB/ChatDev/tree/puppeteer

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MemoryRewardBench: Benchmarking Reward Models for Long-Term Memory Management in Large Language Models

📝 Summary:
MemoryRewardBench is a new benchmark evaluating reward models ability to assess long-term memory management in LLMs across various context lengths and patterns. Evaluations reveal newer RMs outperform predecessors, open-source models are closing the gap, and current RMs have limitations.

🔹 Publication Date: Published on Jan 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11969
• PDF: https://arxiv.org/pdf/2601.11969
• Github: https://github.com/LCM-Lab/MemRewardBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨UniX: Unifying Autoregression and Diffusion for Chest X-Ray Understanding and Generation

📝 Summary:
UniX presents a unified medical foundation model that decouples visual understanding and generation tasks using distinct autoregressive and diffusion branches with cross-modal attention for enhanced p...

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11522
• PDF: https://arxiv.org/pdf/2601.11522
• Github: https://github.com/ZrH42/UniX

🔹 Models citing this paper:
• https://huggingface.co/ZrH42/UniX

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research