✨Self-Improving Pretraining: using post-trained models to pretrain better models

📝 Summary:
A reinforcement learning-based pretraining method improves language model safety, factuality, and quality by evaluating generations through a combination of model rollouts, original suffixes, and rewr...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21343
• PDF: https://arxiv.org/pdf/2601.21343

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Scaling Embeddings Outperforms Scaling Experts in Language Models

📝 Summary:
Embedding scaling offers superior sparsity scaling compared to expert scaling in large language models, enabling efficient inference through system optimizations and speculative decoding. AI-generated...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21204
• PDF: https://arxiv.org/pdf/2601.21204

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DeepSearchQA: Bridging the Comprehensiveness Gap for Deep Research Agents

📝 Summary:
DeepSearchQA presents a 900-prompt benchmark evaluating agents on complex multi-step information-seeking tasks requiring systematic information collation, deduplication, and reasoning about stopping c...

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.20975
• PDF: https://arxiv.org/pdf/2601.20975
• Project Page: https://www.kaggle.com/benchmarks/google/dsqa/leaderboard

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Segment Length Matters: A Study of Segment Lengths on Audio Fingerprinting Performance

📝 Summary:
Neural audio fingerprinting performance varies with segment length, with short segments (0.5-second) generally providing better retrieval accuracy, and large language models showing promise in recomme...

🔹 Publication Date: Published on Jan 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.17690
• PDF: https://arxiv.org/pdf/2601.17690

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨PRISM: Learning Design Knowledge from Data for Stylistic Design Improvement

📝 Summary:
PRISM leverages design data to create a knowledge base for improving graphic designs based on natural language instructions, achieving superior style alignment compared to existing methods. AI-generat...

🔹 Publication Date: Published on Jan 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.11747
• PDF: https://arxiv.org/pdf/2601.11747

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨WorldBench: Disambiguating Physics for Diagnostic Evaluation of World Models

📝 Summary:
WorldBench is introduced as a video-based benchmark for disentangled evaluation of physical reasoning in generative models, revealing specific failure patterns in current state-of-the-art video world ...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21282
• PDF: https://arxiv.org/pdf/2601.21282
• Project Page: https://world-bench.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Idea2Story: An Automated Pipeline for Transforming Research Concepts into Complete Scientific Narratives

📝 Summary:
Offline knowledge construction through structured methodological graphs enables more reliable and scalable autonomous scientific discovery by reducing reliance on real-time literature processing. AI-g...

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.20833
• PDF: https://arxiv.org/pdf/2601.20833

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OCRVerse: Towards Holistic OCR in End-to-End Vision-Language Models

📝 Summary:
OCRVerse unifies text and vision-centric OCR into a holistic end-to-end method for diverse visual documents. It uses comprehensive data and a two-stage SFT-RL training with domain-specific rewards to achieve competitive results.

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21639
• PDF: https://arxiv.org/pdf/2601.21639

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Everything in Its Place: Benchmarking Spatial Intelligence of Text-to-Image Models

📝 Summary:
Text-to-image models struggle with complex spatial reasoning due to sparse prompts. This paper introduces SpatialGenEval, a new benchmark with dense prompts, showing models struggle with higher-order spatial tasks. A new dataset, SpatialT2I, helps fine-tune models for significant performance gain...

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.20354
• PDF: https://arxiv.org/pdf/2601.20354
• Github: https://github.com/AMAP-ML/SpatialGenEval

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

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

#TextToImage #SpatialReasoning #GenerativeAI #ComputerVision #AIResearch

✨MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sources

📝 Summary:
Metric Anything introduces a scalable pretraining framework for metric depth using Sparse Metric Prompts to handle diverse, noisy 3D data. It shows clear scaling trends and achieves state-of-the-art performance across various depth estimation and spatial intelligence tasks.

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22054
• PDF: https://arxiv.org/pdf/2601.22054
• Project Page: https://metric-anything.github.io/metric-anything-io/
• Github: https://github.com/metric-anything/metric-anything

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

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

#MetricDepth #ComputerVision #MachineLearning #DeepLearning #3DVision

✨PLANING: A Loosely Coupled Triangle-Gaussian Framework for Streaming 3D Reconstruction

📝 Summary:
PLANING is an efficient streaming 3D reconstruction framework. It combines explicit geometric primitives and neural Gaussians with decoupled optimization, achieving both high-quality rendering and accurate geometry. It outperforms prior methods in quality and speed.

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22046
• PDF: https://arxiv.org/pdf/2601.22046

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

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

#3DReconstruction #ComputerVision #NeuralNetworks #StreamingTech #ComputerGraphics

✨BMAM: Brain-inspired Multi-Agent Memory Framework

📝 Summary:
BMAM presents a brain-inspired multi-agent memory architecture that decomposes memory into specialized subsystems to address long-term reasoning challenges in language-model-based agents. AI-generated...

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.20465
• PDF: https://arxiv.org/pdf/2601.20465
• Github: https://github.com/innovation64/BMAM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Spotlighting Task-Relevant Features: Object-Centric Representations for Better Generalization in Robotic Manipulation

📝 Summary:
Slot-Based Object-Centric Representations outperform global and dense feature representations in robotic manipulation tasks by providing better generalization under visual distribution shifts. AI-gene...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21416
• PDF: https://arxiv.org/pdf/2601.21416

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨STORM: Slot-based Task-aware Object-centric Representation for robotic Manipulation

📝 Summary:
STORM enhances robotic manipulation by adapting visual foundation models with semantic-aware slots through multi-phase training. This approach improves object discovery, generalization to distractors, and robotic control performance.

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.20381
• PDF: https://arxiv.org/pdf/2601.20381

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

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

#Robotics #AI #ComputerVision #RoboticManipulation #DeepLearning

✨AgentLongBench: A Controllable Long Benchmark For Long-Contexts Agents via Environment Rollouts

📝 Summary:
AgentLongBench evaluates LLM agents via dynamic environment rollouts. It finds agents struggle with high-density tool responses more than memory fragmentation in long conversations, driven by tokens needed to resolve queries.

🔹 Publication Date: Published on Jan 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.20730
• PDF: https://arxiv.org/pdf/2601.20730
• Github: https://github.com/euReKa025/AgentLongBench

✨ Datasets citing this paper:
• https://huggingface.co/datasets/ign1s/AgentLongBench

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

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

#LLMAgents #LongContext #AIResearch #NLP #Benchmarking

✨One-step Latent-free Image Generation with Pixel Mean Flows

📝 Summary:
Pixel MeanFlow pMF proposes a one-step, latent-free image generation method. It separates network output space from loss space, targeting an image manifold for prediction and defining loss in velocity space. pMF achieves strong ImageNet results at 256x256 and 512x512 resolutions.

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22158
• PDF: https://arxiv.org/pdf/2601.22158

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

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

#ImageGeneration #DeepLearning #ComputerVision #GenerativeAI #AIResearch

✨Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts

📝 Summary:
HALO efficiently converts Transformer models to RNN-attention hybrids using minimal training data. This enables superior long-context performance and efficiency, showcased by the HypeNet architecture and its application to the Qwen3 series.

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22156
• PDF: https://arxiv.org/pdf/2601.22156
• Github: https://www.github.com/THUNLP/hybrid-linear-attention

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

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

#HybridAttention #LongContext #Transformers #LLMs #DeepLearning

✨FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning

📝 Summary:
FROST is an attention-aware method that improves reasoning efficiency by pruning uncritical paths and removing reasoning outliers, leading to reduced token usage and improved accuracy. AI-generated su...

🔹 Publication Date: Published on Jan 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.19001
• PDF: https://arxiv.org/pdf/2601.19001

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ECO: Quantized Training without Full-Precision Master Weights

📝 Summary:
Error-compensating optimizer eliminates memory overhead from master weights in quantized LLM training while maintaining near-lossless accuracy. AI-generated summary Quantization has significantly impr...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.22101
• PDF: https://arxiv.org/pdf/2601.22101

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units

📝 Summary:
MDA traces interpretable LLM units to training data using influence functions. Intervening on high-influence samples causally modulates circuit emergence, especially with structural data. This shows a direct link between data, circuit formation, and in-context learning.

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21996
• PDF: https://arxiv.org/pdf/2601.21996
• Github: https://github.com/chenjianhuii/Mechanistic-Data-Attribution

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

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

#LLM #AI #MachineLearning #MechanisticInterpretability #DataAttribution

✨Scalable Power Sampling: Unlocking Efficient, Training-Free Reasoning for LLMs via Distribution Sharpening

📝 Summary:
This paper proposes a training-free method to sharpen LLM distributions, improving reasoning. It approximates the global power distribution with a token-level scaled low-temperature one. This achieves reinforcement learning-like performance with significantly lower computational cost and reduced ...

🔹 Publication Date: Published on Jan 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.21590
• PDF: https://arxiv.org/pdf/2601.21590

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

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

#LLMs #AI #MachineLearning #NLP #DeepLearning