✨Vector Prism: Animating Vector Graphics by Stratifying Semantic Structure

📝 Summary:
A framework aggregates weak predictions to recover semantic structure, enabling coherent SVG animations and improving VLM interactions with vector graphics. AI-generated summary Scalable Vector Graphi...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14336
• PDF: https://arxiv.org/pdf/2512.14336
• Project Page: https://yeolj00.github.io/personal-projects/vector-prism/
• Github: https://github.com/YeolJ00/vector-prism

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EVOLVE-VLA: Test-Time Training from Environment Feedback for Vision-Language-Action Models

📝 Summary:
EVOLVE-VLA, a test-time training framework for Vision-Language-Action models, enables continuous adaptation through environmental interaction with minimal task-specific demonstrations, achieving signi...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14666
• PDF: https://arxiv.org/pdf/2512.14666
• Project Page: https://showlab.github.io/EVOLVE-VLA/
• Github: https://github.com/showlab/EVOLVE-VLA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨TAT: Task-Adaptive Transformer for All-in-One Medical Image Restoration

📝 Summary:
A task-adaptive Transformer (TAT) framework addresses challenges in medical image restoration by dynamically adjusting task-specific weights and loss balances, achieving state-of-the-art performance a...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14550
• PDF: https://arxiv.org/pdf/2512.14550
• Github: https://github.com/Yaziwel/TAT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Zoom-Zero: Reinforced Coarse-to-Fine Video Understanding via Temporal Zoom-in

📝 Summary:
Zoom-Zero, a coarse-to-fine framework, enhances grounded video question answering by improving temporal grounding and answer accuracy through a zoom-in accuracy reward and token-selective credit assig...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14273
• PDF: https://arxiv.org/pdf/2512.14273
• Github: https://xiaoqian-shen.github.io/Zoom-Zero/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MobileWorldBench: Towards Semantic World Modeling For Mobile Agents

📝 Summary:
A novel vision-language model framework improves task success rates for mobile GUI agents by using semantic world models instead of pixel-based predictions. AI-generated summary World models have show...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14014
• PDF: https://arxiv.org/pdf/2512.14014

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨OpenDataArena: A Fair and Open Arena for Benchmarking Post-Training Dataset Value

📝 Summary:
OpenDataArena (ODA) is an open platform that benchmarks post-training datasets for Large Language Models (LLMs) using a unified pipeline, multi-dimensional scoring, and data lineage exploration to enh...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14051
• PDF: https://arxiv.org/pdf/2512.14051
• Project Page: https://opendataarena.github.io
• Github: https://opendataarena.github.io/index.html

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models

📝 Summary:
Cascaded domain-wise reinforcement learning (Cascade RL) is proposed to enhance general-purpose reasoning models, achieving state-of-the-art performance across benchmarks and outperforming the teacher...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13607
• PDF: https://arxiv.org/pdf/2512.13607

🔹 Models citing this paper:
• https://huggingface.co/nvidia/Nemotron-Cascade-8B
• https://huggingface.co/nvidia/Nemotron-Cascade-14B-Thinking
• https://huggingface.co/nvidia/Nemotron-Cascade-8B-Thinking

✨ Datasets citing this paper:
• https://huggingface.co/datasets/nvidia/Nemotron-Cascade-RL-SWE
• https://huggingface.co/datasets/nvidia/Nemotron-Cascade-SFT-SWE

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨S2D: Sparse-To-Dense Keymask Distillation for Unsupervised Video Instance Segmentation

📝 Summary:
An unsupervised video instance segmentation model using real video data and deep motion priors outperforms existing methods by establishing temporal coherence and using sparse-to-dense distillation. A...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14440
• PDF: https://arxiv.org/pdf/2512.14440
• Project Page: https://leonsick.github.io/s2d
• Github: https://github.com/leonsick/s2d

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MeViS: A Multi-Modal Dataset for Referring Motion Expression Video Segmentation

📝 Summary:
MeViS is a multi-modal dataset for referring motion expression video segmentation, addressing the need to segment and track objects based on their motion denoscriptions. It provides text and audio annotations for complex videos, enabling research into motion-guided video understanding.

🔹 Publication Date: Published on Dec 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.10945
• PDF: https://arxiv.org/pdf/2512.10945
• Project Page: https://henghuiding.com/MeViS/

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

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

#VideoSegmentation #MultiModalAI #ComputerVision #Dataset #MotionUnderstanding

✨Spherical Leech Quantization for Visual Tokenization and Generation

📝 Summary:
This paper presents Spherical Leech Quantization L24-SQ, a Leech lattice-based method. It simplifies training and improves reconstruction-compression for image tokenization, compression, and generation, outperforming prior art like BSQ.

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14697
• PDF: https://arxiv.org/pdf/2512.14697
• Project Page: https://cs.stanford.edu/~yzz/npq/
• Github: https://github.com/zhaoyue-zephyrus/InfinityCC

🔹 Models citing this paper:
• https://huggingface.co/zhaoyue-zephyrus/InfinityCC_L24SQ

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨JMMMU-Pro: Image-based Japanese Multi-discipline Multimodal Understanding Benchmark via Vibe Benchmark Construction

📝 Summary:
JMMMU-Pro, an image-based Japanese multi-discipline multimodal understanding benchmark, challenges open-source large multimodal models through integrated visual-textual understanding and is constructe...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2512.14620
• PDF: https://arxiv.org/pdf/2512.14620
• Github: https://mmmu-japanese-benchmark.github.io/JMMMU_Pro/

✨ Datasets citing this paper:
• https://huggingface.co/datasets/JMMMU/JMMMU-Pro

✨ Spaces citing this paper:
• https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Reveal Hidden Pitfalls and Navigate Next Generation of Vector Similarity Search from Task-Centric Views

📝 Summary:
Iceberg is a new benchmark for vector similarity search VSS that evaluates methods from a task-centric view. It uncovers performance degradation, re-ranks VSS algorithms based on application-level metrics, and guides practitioners.

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.12980
• PDF: https://arxiv.org/pdf/2512.12980

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

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

#VectorSimilaritySearch #MachineLearning #DataScience #Benchmarks #Algorithms

✨Janus: Disaggregating Attention and Experts for Scalable MoE Inference

📝 Summary:
Janus is a scalable MoE inference system that disaggregates attention and expert modules onto separate GPU sub-clusters. This enables independent scaling for each module, improving throughput and latency by optimizing resource utilization.

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2512.13525
• PDF: https://arxiv.org/pdf/2512.13525

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Differentiable Evolutionary Reinforcement Learning

📝 Summary:
A bilevel differentiable approach for evolving reward functions in reinforcement learning enhances agent performance across various domains by capturing task structure through reinforcement learning. ...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13399
• PDF: https://arxiv.org/pdf/2512.13399
• Github: https://github.com/sitaocheng/DERL

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ContextAnyone: Context-Aware Diffusion for Character-Consistent Text-to-Video Generation

📝 Summary:
ContextAnyone uses a diffusion framework with emphasis attention and gap-rope positional embeddings to generate consistent character videos from text and a single reference image. AI-generated summary...

🔹 Publication Date: Published on Dec 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.07328
• PDF: https://arxiv.org/pdf/2512.07328

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research