✨Skyra: AI-Generated Video Detection via Grounded Artifact Reasoning

📝 Summary:
Skyra, a specialized multimodal large language model, detects and explains visual artifacts in AI-generated videos using a novel dataset and two-stage training strategy, outperforming existing methods...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15693
• PDF: https://arxiv.org/pdf/2512.15693
• Project Page: https://joeleelyf.github.io/Skyra/
• Github: https://github.com/JoeLeelyf/Skyra

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Fast and Accurate Causal Parallel Decoding using Jacobi Forcing

📝 Summary:
Jacobi Forcing is a progressive distillation method that enables efficient parallel decoding of transformer-based models while maintaining performance, significantly reducing inference latency. AI-gen...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14681
• PDF: https://arxiv.org/pdf/2512.14681
• Github: https://github.com/hao-ai-lab/JacobiForcing

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DiffusionVL: Translating Any Autoregressive Models into Diffusion Vision Language Models

📝 Summary:
DiffusionVL, a family of diffusion vision language models derived from autoregressive models through fine-tuning, achieves performance improvements and faster inference speeds compared to existing mod...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15713
• PDF: https://arxiv.org/pdf/2512.15713

🔹 Models citing this paper:
• https://huggingface.co/hustvl/DiffusionVL-Qwen2.5VL-3B
• https://huggingface.co/hustvl/DiffusionVL-Qwen2.5VL-7B

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Qwen-Image-Layered: Towards Inherent Editability via Layer Decomposition

📝 Summary:
Qwen-Image-Layered decomposes images into semantically disentangled RGBA layers using a diffusion model, enabling independent editing of each layer and improving decomposition quality and consistency....

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15603
• PDF: https://arxiv.org/pdf/2512.15603

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Step-GUI Technical Report

📝 Summary:
A self-evolving training pipeline with the Calibrated Step Reward System and GUI-MCP protocol improve GUI automation efficiency, accuracy, and privacy in real-world scenarios. AI-generated summary Rec...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15431
• PDF: https://arxiv.org/pdf/2512.15431

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Robust and Calibrated Detection of Authentic Multimedia Content

📝 Summary:
A resynthesis framework enhances deepfake detection by verifying authenticity with low false positive rates and robustness against efficient adversaries, supporting multiple modalities. AI-generated s...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15182
• PDF: https://arxiv.org/pdf/2512.15182

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Is Nano Banana Pro a Low-Level Vision All-Rounder? A Comprehensive Evaluation on 14 Tasks and 40 Datasets

📝 Summary:
Nano Banana Pro excels in subjective visual quality across low-level vision tasks without fine-tuning but struggles with traditional reference-based quantitative metrics due to generative model stocha...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15110
• PDF: https://arxiv.org/pdf/2512.15110
• Project Page: https://lowlevelbanana.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning

📝 Summary:
The paper proposes SAGE, a multi-turn reasoning system for video that mimics human behavior, using synthetic data and reinforcement learning to improve performance on long videos. AI-generated summary...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13874
• PDF: https://arxiv.org/pdf/2512.13874
• Project Page: https://praeclarumjj3.github.io/sage/
• Github: https://github.com/allenai/SAGE

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨In Pursuit of Pixel Supervision for Visual Pre-training

📝 Summary:
Pixio, an enhanced masked autoencoder, demonstrates competitive performance across various downstream tasks using pixel-space self-supervised learning, outperforming latent-space approaches. AI-genera...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15715
• PDF: https://arxiv.org/pdf/2512.15715
• Project Page: https://github.com/facebookresearch/pixio
• Github: https://github.com/facebookresearch/pixio

🔹 Models citing this paper:
• https://huggingface.co/facebook/pixio-vitb16
• https://huggingface.co/facebook/pixio-vitl16
• https://huggingface.co/facebook/pixio-vit1b16

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨End-to-End Training for Autoregressive Video Diffusion via Self-Resampling

📝 Summary:
Resampling Forcing is a teacher-free framework to train autoregressive video diffusion models. It uses self-resampling to simulate inference errors and history routing for efficient long video generation. This approach improves temporal consistency and achieves comparable performance to teacher-b...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15702
• PDF: https://arxiv.org/pdf/2512.15702

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨LikeBench: Evaluating Subjective Likability in LLMs for Personalization

📝 Summary:
LikeBench introduces a multi-session evaluation framework to measure the likability of LLMs by their ability to adapt to user preferences across multiple dimensions, demonstrating that strong memory p...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13077
• PDF: https://arxiv.org/pdf/2512.13077

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨IC-Effect: Precise and Efficient Video Effects Editing via In-Context Learning

📝 Summary:
IC-Effect is an instruction-guided DiT framework for precise video VFX editing. It synthesizes complex effects with spatial-temporal consistency by leveraging contextual learning, a two-stage training strategy, and sparse tokenization, outperforming existing models.

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15635
• PDF: https://arxiv.org/pdf/2512.15635

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨WAY: Estimation of Vessel Destination in Worldwide AIS Trajectory

📝 Summary:
A novel deep learning architecture, WAY, uses nested sequence structures and spatial grids for accurate long-term vessel destination estimation from AIS data, incorporating CASP blocks and Gradient Dr...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13190
• PDF: https://arxiv.org/pdf/2512.13190
• Github: https://github.com/sadPororo/WAY

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MMSI-Video-Bench: A Holistic Benchmark for Video-Based Spatial Intelligence

📝 Summary:
MMSI-Video-Bench is a comprehensive benchmark for video-based spatial intelligence in MLLMs, revealing significant gaps between human and AI performance and highlighting challenges in geometric reason...

🔹 Publication Date: Published on Dec 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.10863
• PDF: https://arxiv.org/pdf/2512.10863
• Github: https://github.com/InternRobotics/MMSI-Video-Bench

✨ Datasets citing this paper:
• https://huggingface.co/datasets/rbler/MMSI-Video-Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Hybrid Attribution Priors for Explainable and Robust Model Training

📝 Summary:
A novel framework, Class-Aware Attribution Prior (CAP), enhances language model interpretability and robustness by guiding the model to capture fine-grained class distinctions and combining with exist...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14719
• PDF: https://arxiv.org/pdf/2512.14719

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SS4D: Native 4D Generative Model via Structured Spacetime Latents

📝 Summary:
SS4D synthesizes dynamic 3D objects from monocular video using a native 4D generative model with structured spacetime latents, ensuring high fidelity, temporal coherence, and structural consistency. A...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14284
• PDF: https://arxiv.org/pdf/2512.14284
• Project Page: https://lizb6626.github.io/SS4D/
• Github: https://github.com/Lizb6626/SS4D/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨VOYAGER: A Training Free Approach for Generating Diverse Datasets using LLMs

📝 Summary:
Voyager is a novel, training-free method that iteratively generates diverse synthetic datasets from LLMs. It uses determinantal point processes to optimize diversity, significantly outperforming baselines with a 1.5-3x improvement.

🔹 Publication Date: Published on Dec 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.12072
• PDF: https://arxiv.org/pdf/2512.12072

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

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

#LLMs #SyntheticData #DataScience #MachineLearning #AI

✨HyperVL: An Efficient and Dynamic Multimodal Large Language Model for Edge Devices

📝 Summary:
HyperVL is an efficient multimodal large language model for edge devices. It uses image tiling, a Visual Resolution Compressor, and Dual Consistency Learning to reduce memory, latency, and power. HyperVL maintains performance, making it practical for on-device inference.

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14052
• PDF: https://arxiv.org/pdf/2512.14052

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

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

#HyperVL #MLLM #EdgeAI #EfficientAI #OnDeviceAI

✨Towards Seamless Interaction: Causal Turn-Level Modeling of Interactive 3D Conversational Head Dynamics

📝 Summary:
TIMAR is a new causal framework for 3D conversational head generation. It models dialogue using interleaved audio-visual contexts to predict continuous head dynamics, improving coherence and expressive variability. Experiments show TIMAR significantly reduces errors and improves performance.

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15340
• PDF: https://arxiv.org/pdf/2512.15340
• Project Page: https://github.com/CoderChen01/towards-seamleass-interaction/blob/main/README.md
• Github: https://github.com/CoderChen01/towards-seamleass-interaction

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

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

#ConversationalAI #3DAnimation #HumanComputerInteraction #CausalModeling #AI