✨ChronosObserver: Taming 4D World with Hyperspace Diffusion Sampling

📝 Summary:
ChronosObserver generates high-fidelity, 3D-consistent, and time-synchronized multi-view videos. It is a training-free method leveraging World State Hyperspace and Hyperspace Guided Sampling to synchronize views. This approach overcomes challenges in 4D world generation without model training.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01481
• PDF: https://arxiv.org/pdf/2512.01481
• Project Page: https://icvteam.github.io/ChronosObserver.html

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#4DGeneration #DiffusionModels #ComputerVision #MultiViewVideo #AIResearch

✨WiseEdit: Benchmarking Cognition- and Creativity-Informed Image Editing

📝 Summary:
WiseEdit is a new benchmark for evaluating image editing models, focusing on cognition and creativity. It decomposes editing into Awareness, Interpretation, and Imagination tasks, assessing declarative, procedural, and metacognitive knowledge. This reveals limitations in current models.

🔹 Publication Date: Published on Nov 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00387
• PDF: https://arxiv.org/pdf/2512.00387
• Project Page: https://qnancy.github.io/wiseedit_project_page/

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#ImageEditing #ComputerVision #AIResearch #CognitiveAI #CreativeAI

✨IndicParam: Benchmark to evaluate LLMs on low-resource Indic Languages

📝 Summary:
IndicParam is a new benchmark with over 13000 multiple-choice questions for 11 low-resource Indic languages. It reveals that even top LLMs achieve only ~45% accuracy, showing limitations in cross-lingual transfer and grammatical proficiency. The benchmark also assesses diverse question formats.

🔹 Publication Date: Published on Nov 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00333
• PDF: https://arxiv.org/pdf/2512.00333
• Project Page: https://huggingface.co/datasets/bharatgenai/IndicParam
• Github: https://github.com/ayushbits/IndicParam

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#LLM #NLP #LowResourceLanguages #IndicLanguages #AIResearch

✨From Code Foundation Models to Agents and Applications: A Practical Guide to Code Intelligence

📝 Summary:
This paper provides a practical guide to code LLMs, covering their lifecycle from data to deployment. It examines techniques, analyzes various models, and discusses real-world challenges like correctness and security. Experiments on pre-training and fine-tuning are included.

🔹 Publication Date: Published on Nov 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.18538
• PDF: https://arxiv.org/pdf/2511.18538

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#CodeLLMs #AI #MachineLearning #SoftwareEngineering #FoundationModels

✨Stabilizing Reinforcement Learning with LLMs: Formulation and Practices

📝 Summary:
This paper presents a theoretical foundation for stabilizing RL with LLMs, optimizing sequence rewards via token-level objectives. It highlights that importance sampling, clipping, and Routing Replay minimize policy staleness, crucial for stable training. Stabilized training consistently yields c...

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01374
• PDF: https://arxiv.org/pdf/2512.01374

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#ReinforcementLearning #LLMs #AI #MachineLearning #AIResearch

✨PromptBridge: Cross-Model Prompt Transfer for Large Language Models

📝 Summary:
PromptBridge combats Model Drifting, where prompts lose effectiveness across LLMs. Training-free, it enables cross-model prompt transfer by mapping source prompts to optimized target prompts, improving accuracy, reducing re-optimization.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01420
• PDF: https://arxiv.org/pdf/2512.01420

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#LLM #PromptEngineering #AI #ModelDrifting #PromptBridge

✨The Art of Scaling Test-Time Compute for Large Language Models

📝 Summary:
This study systematically compares Test-Time Scaling TTS strategies for LLMs. It finds no single dominant strategy, identifies distinct model trace-quality patterns, and shows optimal performance scales with compute. A practical guide for selecting TTS strategies is provided.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2512.02008
• PDF: https://arxiv.org/pdf/2512.02008
• Github: https://github.com/Aradhye2002/art_of_tts

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#LLM #TestTimeScaling #AI #DeepLearning #NLP

✨StreamGaze: Gaze-Guided Temporal Reasoning and Proactive Understanding in Streaming Videos

📝 Summary:
StreamGaze is a new benchmark evaluating how MLLMs use human gaze for temporal and proactive reasoning in streaming videos. It reveals significant performance gaps between current AI models and human abilities in gaze-based temporal reasoning and proactive prediction.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01707
• PDF: https://arxiv.org/pdf/2512.01707
• Project Page: https://streamgaze.github.io/
• Github: https://github.com/daeunni/StreamGaze

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#StreamGaze #MLLMs #TemporalReasoning #ComputerVision #AI

✨Asking like Socrates: Socrates helps VLMs understand remote sensing images

📝 Summary:
Remote sensing models often show fake reasoning from coarse image understanding. This paper introduces RS-EoT, an iterative, language-driven system with a Socratic multi-agent approach and RL to seek visual evidence. It achieves state-of-the-art results, enabling genuine, evidence-grounded reason...

🔹 Publication Date: Published on Nov 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.22396
• PDF: https://arxiv.org/pdf/2511.22396
• Project Page: https://geox-lab.github.io/Asking_like_Socrates/
• Github: https://github.com/GeoX-Lab/Asking_like_Socrates

🔹 Models citing this paper:
• https://huggingface.co/ShaoRun/RS-EoT-7B

✨ Datasets citing this paper:
• https://huggingface.co/datasets/ShaoRun/RS-EoT-4K

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#VLM #RemoteSensing #AI #ReinforcementLearning #MultiAgentSystems

✨Accelerating Streaming Video Large Language Models via Hierarchical Token Compression

📝 Summary:
Streaming VideoLLMs face high latency from ViT encoding and LLM pre-filling. STC, a hierarchical framework, optimizes this by caching features and pruning tokens. It reduces latency by up to 24.5 percent for ViT and 45.3 percent for LLM pre-filling, retaining 99 percent accuracy.

🔹 Publication Date: Published on Nov 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00891
• PDF: https://arxiv.org/pdf/2512.00891
• Github: https://github.com/lern-to-write/STC

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#VideoLLM #LLM #DeepLearning #AI #PerformanceOptimization

✨Script: Graph-Structured and Query-Conditioned Semantic Token Pruning for Multimodal Large Language Models

📝 Summary:
Script is a new plug-and-play token pruning method for multimodal large language models. It uses graph-structured and query-conditioned modules to remove redundant visual tokens while preserving relevant information without retraining. This boosts efficiency and accuracy, achieving significant sp...

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01949
• PDF: https://arxiv.org/pdf/2512.01949
• Github: https://01yzzyu.github.io/noscript.github.io/

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#MultimodalAI #LLMs #TokenPruning #DeepLearning #Efficiency

✨OpenREAD: Reinforced Open-Ended Reasoing for End-to-End Autonomous Driving with LLM-as-Critic

📝 Summary:
OpenREAD enhances autonomous driving via end-to-end reinforcement fine-tuning for both reasoning and planning. It uses an LLM critic to quantify open-ended reasoning, achieving state-of-the-art performance by addressing prior limitations.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01830
• PDF: https://arxiv.org/pdf/2512.01830
• Github: https://github.com/wyddmw/OpenREAD

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#AutonomousDriving #LLMs #ReinforcementLearning #AI #Robotics

✨Flash-DMD: Towards High-Fidelity Few-Step Image Generation with Efficient Distillation and Joint Reinforcement Learning

📝 Summary:
Flash-DMD accelerates generative diffusion models via efficient timestep-aware distillation and joint reinforcement learning. This framework achieves faster convergence, high-fidelity few-step generation, and stabilizes RL training using distillation as a regularizer, all with reduced computation...

🔹 Publication Date: Published on Nov 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.20549
• PDF: https://arxiv.org/pdf/2511.20549

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#DiffusionModels #ImageGeneration #ReinforcementLearning #ModelDistillation #GenerativeAI

✨OmniFusion: Simultaneous Multilingual Multimodal Translations via Modular Fusion

📝 Summary:
OmniFusion is a multimodal translation system integrating pretrained foundation models with LLMs via a novel fusion strategy. It enables simultaneous multilingual translation using audio and visual inputs, reducing latency and improving quality over cascaded systems.

🔹 Publication Date: Published on Nov 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00234
• PDF: https://arxiv.org/pdf/2512.00234
• Github: https://github.com/saikoneru/OmniFusion

🔹 Models citing this paper:
• https://huggingface.co/skoneru/OmniFusion
• https://huggingface.co/skoneru/OmniFusion_v2

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#MultimodalAI #LLMs #MachineTranslation #FoundationModels #AIResearch

✨Structured Extraction from Business Process Diagrams Using Vision-Language Models

📝 Summary:
This paper presents a method using Vision-Language Models to extract structured JSON from BPMN diagram images. It incorporates OCR for text enrichment, demonstrating improved model performance and enabling extraction when source files are unavailable.

🔹 Publication Date: Published on Nov 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.22448
• PDF: https://arxiv.org/pdf/2511.22448
• Github: https://github.com/pritamdeka/BPMN-VLM

✨ Datasets citing this paper:
• https://huggingface.co/datasets/pritamdeka/BPMN-VLM

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#VisionLanguageModels #BPMN #InformationExtraction #AI #ComputerVision

✨LFM2 Technical Report

📝 Summary:
LFM2 is a family of compact foundation models designed for efficient on-device deployment. It uses hardware-in-the-loop architecture search and advanced training to achieve high performance across diverse tasks, including multimodal applications.

🔹 Publication Date: Published on Nov 28

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.23404
• PDF: https://arxiv.org/pdf/2511.23404

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#FoundationModels #EdgeAI #MultimodalAI #AIResearch #MachineLearning

✨Learning Eigenstructures of Unstructured Data Manifolds

📝 Summary:
This deep learning framework learns spectral bases directly from unstructured data, eliminating traditional operator selection and eigendecomposition. It provides a data-driven alternative for geometry processing, recovering spectral bases and eigenvalues unsupervised without explicit operator co...

🔹 Publication Date: Published on Nov 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01103
• PDF: https://arxiv.org/pdf/2512.01103

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#DeepLearning #DataScience #ManifoldLearning #GeometryProcessing #UnsupervisedLearning

✨Wikontic: Constructing Wikidata-Aligned, Ontology-Aware Knowledge Graphs with Large Language Models

📝 Summary:
Wikontic is a multi-stage pipeline that builds high-quality, ontology-consistent knowledge graphs from text. It achieves state-of-the-art performance in information retention and efficiency, providing structured grounding for LLMs.

🔹 Publication Date: Published on Nov 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00590
• PDF: https://arxiv.org/pdf/2512.00590

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#KnowledgeGraphs #LLMs #Ontologies #NLP #AI

✨SCALE: Selective Resource Allocation for Overcoming Performance Bottlenecks in Mathematical Test-time Scaling

📝 Summary:
SCALE improves LLM math reasoning by selectively allocating resources based on sub-problem difficulty. It addresses uniform allocation bottlenecks, boosting accuracy up to 13.75% and cutting costs by 33-53% compared to uniform scaling.

🔹 Publication Date: Published on Nov 29

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.00466
• PDF: https://arxiv.org/pdf/2512.00466
• Github: https://github.com/XiaoYang66/DualThinking

✨ Datasets citing this paper:
• https://huggingface.co/datasets/YangXiao-nlp/DualThinking

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#LLM #AI #MachineLearning #PerformanceOptimization #MathReasoning

✨Agentic Policy Optimization via Instruction-Policy Co-Evolution

📝 Summary:
INSPO introduces a novel framework dynamically optimizing instructions within the reinforcement learning loop for autonomous agents. It substantially outperforms static instruction methods in multi-turn reasoning by discovering innovative, strategic reasoning paths.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01945
• PDF: https://arxiv.org/pdf/2512.01945
• Github: https://github.com/cambridgeltl/inspo

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#ReinforcementLearning #AIAgents #PolicyOptimization #MachineLearning #AI

✨Where Culture Fades: Revealing the Cultural Gap in Text-to-Image Generation

📝 Summary:
Multilingual text-to-image models often generate culturally neutral images. This paper identifies specific neurons for cultural information and proposes two strategies: inference-time activation and layer-targeted enhancement. These methods improve cultural consistency while preserving image qual...

🔹 Publication Date: Published on Nov 21

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2511.17282
• PDF: https://arxiv.org/pdf/2511.17282

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

For more data science resources:
✓ https://news.1rj.ru/str/DataScienceT

#TextToImage #CulturalAI #ResponsibleAI #DeepLearning #AIResearch