✨Terrain Diffusion: A Diffusion-Based Successor to Perlin Noise in Infinite, Real-Time Terrain Generation

📝 Summary:
Terrain Diffusion uses diffusion models and a novel algorithm called InfiniteDiffusion to generate realistic, seamless, and boundless procedural worlds with constant-time random access. AI-generated s...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.08309
• PDF: https://arxiv.org/pdf/2512.08309
• Project Page: https://xandergos.github.io/terrain-diffusion/
• Github: https://github.com/xandergos/terrain-diffusion

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨See, Hear, and Understand: Benchmarking Audiovisual Human Speech Understanding in Multimodal Large Language Models

📝 Summary:
AV-SpeakerBench is a new benchmark assessing speaker-centric audiovisual reasoning in MLLMs. It features 3,212 expert-curated questions focused on precise speech understanding. Gemini models outperform open-source systems, particularly in audiovisual fusion capabilities.

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.02231
• PDF: https://arxiv.org/pdf/2512.02231
• Project Page: https://plnguyen2908.github.io/AV-SpeakerBench-project-page/
• Github: https://github.com/plnguyen2908/AV-SpeakerBench

✨ Datasets citing this paper:
• https://huggingface.co/datasets/plnguyen2908/AV-SpeakerBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Arbitrage: Efficient Reasoning via Advantage-Aware Speculation

📝 Summary:
Arbitrage is a speculative decoding framework for LLMs that dynamically routes generation. It uses a router to predict when a target model will provide a better reasoning step, preventing wasted compute from regenerating rejected steps. This approach reduces inference latency by up to two times w...

🔹 Publication Date: Published on Dec 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05033
• PDF: https://arxiv.org/pdf/2512.05033
• Project Page: https://www.monishwaran.com/arbitrage.html
• Github: https://github.com/SqueezeAILab/Arbitrage

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨COREA: Coarse-to-Fine 3D Representation Alignment Between Relightable 3D Gaussians and SDF via Bidirectional 3D-to-3D Supervision

📝 Summary:
COREA unifies 3D Gaussians and SDF for accurate geometry and relighting. It uses a coarse-to-fine bidirectional 3D-to-3D alignment, learning geometry directly in 3D to overcome prior limitations. This improves novel-view synthesis, mesh reconstruction, and physically-based rendering.

🔹 Publication Date: Published on Dec 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.07107
• PDF: https://arxiv.org/pdf/2512.07107
• Project Page: https://cau-vilab.github.io/COREA/
• Github: https://github.com/CAU-VILab/COREA-arXiv

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SegEarth-OV3: Exploring SAM 3 for Open-Vocabulary Semantic Segmentation in Remote Sensing Images

📝 Summary:
A preliminary exploration of using SAM 3 for remote sensing open-vocabulary semantic segmentation demonstrates promising results through a mask fusion strategy and presence score filtering. AI-generat...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.08730
• PDF: https://arxiv.org/pdf/2512.08730
• Github: https://github.com/earth-insights/SegEarth-OV-3

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨UniUGP: Unifying Understanding, Generation, and Planing For End-to-end Autonomous Driving

📝 Summary:
A unified framework combines vision-language models and video generation to improve autonomous driving in complex scenarios by enhancing reasoning, trajectory planning, and video generation. AI-genera...

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09864
• PDF: https://arxiv.org/pdf/2512.09864

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨StereoWorld: Geometry-Aware Monocular-to-Stereo Video Generation

📝 Summary:
StereoWorld generates high-quality stereo video from monocular input using a pretrained video generator with geometry-aware regularization and spatio-temporal tiling. AI-generated summary The growing ...

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09363
• PDF: https://arxiv.org/pdf/2512.09363

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OmniPSD: Layered PSD Generation with Diffusion Transformer

📝 Summary:
OmniPSD, a diffusion framework within the Flux ecosystem, enables text-to-PSD generation and image-to-PSD decomposition, achieving high-fidelity results with transparency awareness. AI-generated summa...

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09247
• PDF: https://arxiv.org/pdf/2512.09247

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨WonderZoom: Multi-Scale 3D World Generation

📝 Summary:
WonderZoom generates multi-scale 3D scenes from a single image using scale-adaptive Gaussian surfels and a progressive detail synthesizer, outperforming existing models in quality and alignment. AI-ge...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09164
• PDF: https://arxiv.org/pdf/2512.09164

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learning Unmasking Policies for Diffusion Language Models

📝 Summary:
Reinforcement learning is used to train sampling procedures for masked discrete diffusion language models, improving token throughput and quality compared to heuristic strategies. AI-generated summary...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09106
• PDF: https://arxiv.org/pdf/2512.09106

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Reinventing Clinical Dialogue: Agentic Paradigms for LLM Enabled Healthcare Communication

📝 Summary:
The survey analyzes the cognitive architecture of medical AI systems, focusing on the shift from generative text prediction to agentic autonomy, and categorizes methods into four archetypes based on k...

🔹 Publication Date: Published on Dec 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.01453
• PDF: https://arxiv.org/pdf/2512.01453

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨HiF-VLA: Hindsight, Insight and Foresight through Motion Representation for Vision-Language-Action Models

📝 Summary:
HiF-VLA improves long-horizon robotic manipulation by using motion for bidirectional temporal reasoning. It addresses VLA model temporal myopia by integrating past dynamics hindsight and anticipating future motion foresight. This framework significantly outperforms baselines with negligible latency.

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09928
• PDF: https://arxiv.org/pdf/2512.09928
• Project Page: https://github.com/OpenHelix-Team/HiF-VLA
• Github: https://hifvla.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Pay Less Attention to Function Words for Free Robustness of Vision-Language Models

📝 Summary:
Function-word De-Attention (FDA) mitigates adversarial attacks on robust VLMs by differentially subtracting function-word cross-attention, improving robustness with minimal performance trade-offs. AI-...

🔹 Publication Date: Published on Dec 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.07222
• PDF: https://arxiv.org/pdf/2512.07222
• Github: https://github.com/michaeltian108/FDA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules

📝 Summary:
SchED, a training-free early-exit algorithm, accelerates diffusion large language model decoding with minimal performance loss across various tasks. AI-generated summary Diffusion large language model...

🔹 Publication Date: Published on Dec 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.02892
• PDF: https://arxiv.org/pdf/2512.02892

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨InfiniteVL: Synergizing Linear and Sparse Attention for Highly-Efficient, Unlimited-Input Vision-Language Models

📝 Summary:
InfiniteVL is a linear-complexity VLM architecture combining sliding window attention and Gated DeltaNet. It surpasses prior linear models and matches leading Transformers with less data, achieving over 3.6 times faster inference and robust long-term memory.

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.08829
• PDF: https://arxiv.org/pdf/2512.08829
• Project Page: https://github.com/hustvl/InfiniteVL
• Github: https://github.com/hustvl/InfiniteVL

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EtCon: Edit-then-Consolidate for Reliable Knowledge Editing

📝 Summary:
A novel knowledge editing framework, Edit-then-Consolidate, addresses overfitting and lack of knowledge integration in large language models through targeted fine-tuning and policy optimization, enhan...

🔹 Publication Date: Published on Dec 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04753
• PDF: https://arxiv.org/pdf/2512.04753
• Github: https://github.com/RlinL/EtCon

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Beyond Unified Models: A Service-Oriented Approach to Low Latency, Context Aware Phonemization for Real Time TTS

📝 Summary:
A framework is proposed to improve phonemization quality in TTS systems without sacrificing real-time performance through lightweight context-aware phonemization and a service-oriented architecture. A...

🔹 Publication Date: Published on Dec 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.08006
• PDF: https://arxiv.org/pdf/2512.08006
• Github: https://github.com/MahtaFetrat/Piper-with-LCA-Phonemizer

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Composing Concepts from Images and Videos via Concept-prompt Binding

📝 Summary:
Bind & Compose introduces a one-shot method for composing visual concepts from images and videos. It binds concepts to prompt tokens using hierarchical binders and novel strategies, achieving superior consistency, fidelity, and motion quality.

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09824
• PDF: https://arxiv.org/pdf/2512.09824
• Project Page: https://refkxh.github.io/BiCo_Webpage/
• Github: https://github.com/refkxh/bico

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

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

#AI #ComputerVision #GenerativeAI #VisualConcepts #PromptEngineering

✨TED-4DGS: Temporally Activated and Embedding-based Deformation for 4DGS Compression

📝 Summary:
TED-4DGS efficiently compresses dynamic 3D scenes using sparse anchor-based 3D Gaussian Splatting with novel temporal activation and embedding-based deformation. It optimizes rate-distortion with an implicit neural representation hyperprior and autoregressive model, achieving state-of-the-art com...

🔹 Publication Date: Published on Dec 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.05446
• PDF: https://arxiv.org/pdf/2512.05446

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

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

#4DGS #3DCompression #NeuralRendering #ComputerVision #DynamicScenes

✨VideoSSM: Autoregressive Long Video Generation with Hybrid State-Space Memory

📝 Summary:
VideoSSM proposes a hybrid state-space memory model for long video generation. It unifies autoregressive diffusion with global state-space memory and local context to achieve state-of-the-art temporal consistency and motion stability. This enables scalable, interactive minute-scale video synthesis.

🔹 Publication Date: Published on Dec 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.04519
• PDF: https://arxiv.org/pdf/2512.04519

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

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

#VideoGeneration #GenerativeAI #DiffusionModels #StateSpaceModels #DeepLearning

✨BrainExplore: Large-Scale Discovery of Interpretable Visual Representations in the Human Brain

📝 Summary:
An automated framework identifies and explains visual representations in human brain fMRI data using unsupervised decomposition and natural language denoscriptions. This large-scale method reveals thousands of interpretable visual concepts, including previously unknown fine-grained representations ...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.08560
• PDF: https://arxiv.org/pdf/2512.08560
• Project Page: https://navvewas.github.io/BrainExplore/

✨ Spaces citing this paper:
• https://huggingface.co/spaces/mcosarinsky/BrainExplore-demo

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

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

#Neuroscience #BrainMapping #fMRI #AIResearch #DataScience