✨Puzzle Curriculum GRPO for Vision-Centric Reasoning

📝 Summary:
Puzzle Curriculum GRPO PC-GRPO improves VLM visual reasoning without annotations. It uses self-supervised puzzle environments for verifiable rewards and a difficulty-aware curriculum to enhance consistency and accuracy.

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14944
• PDF: https://arxiv.org/pdf/2512.14944
• Project Page: https://pcgrpo.github.io/

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

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

#VLM #VisualReasoning #SelfSupervisedLearning #ComputerVision #AI

✨FrontierCS: Evolving Challenges for Evolving Intelligence

📝 Summary:
FrontierCS is a new benchmark for evaluating models on 156 open-ended computer science problems with unknown optimal solutions. Models must implement executable programs for tasks like NP-hard algorithmic and research problems. Empirical results show models lag human experts and over-optimize for...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15699
• PDF: https://arxiv.org/pdf/2512.15699
• Github: https://github.com/FrontierCS/Frontier-CS

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

“I spent hours lost in endless Telegram groups—until I discovered this hidden search engine.” Argo🔍Search lets you find the best groups, channels, music, and news in seconds. No more wasting time scrolling!

Discover what others haven’t yet: Try it now and unlock Telegram like never before.

#ad InsideAds

✨SonicMoE: Accelerating MoE with IO and Tile-aware Optimizations

📝 Summary:
SonicMoE optimizes Mixture of Experts models by reducing activation memory through minimal caching, overlapping IO with computation, and using token rounding to minimize padding waste. This achieves a 45 percent memory reduction and significantly improves compute throughput, enabling faster MoE t...

🔹 Publication Date: Published on Dec 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.14080
• PDF: https://arxiv.org/pdf/2512.14080

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Kling-Omni Technical Report

📝 Summary:
Kling-Omni is a versatile generative framework that synthesizes high-quality videos from multimodal inputs. It unifies video generation, editing, and reasoning tasks, supporting diverse inputs to create cinematic content. This system represents a pivotal step toward multimodal world simulators.

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16776
• PDF: https://arxiv.org/pdf/2512.16776

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨LLaDA2.0: Scaling Up Diffusion Language Models to 100B

📝 Summary:
LLaDA2.0 converts auto-regressive models into discrete diffusion large language models using a block-level training scheme, improving efficiency and performance at large scales. AI-generated summary T...

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15745
• PDF: https://arxiv.org/pdf/2512.15745

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DeContext as Defense: Safe Image Editing in Diffusion Transformers

📝 Summary:
DeContext defends against unauthorized in-context image editing by weakening cross-attention pathways in multimodal attention layers, preserving visual quality while blocking unwanted modifications. A...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16625
• PDF: https://arxiv.org/pdf/2512.16625
• Project Page: https://linghuiishen.github.io/decontext_project_page/
• Github: https://github.com/LinghuiiShen/DeContext

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Adaptation of Agentic AI

📝 Summary:
This paper presents a framework for agent and tool adaptation in agentic AI systems, clarifying design strategies and identifying open challenges for improving AI capabilities. AI-generated summary Cu...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16301
• PDF: https://arxiv.org/pdf/2512.16301
• Github: https://github.com/pat-jj/Awesome-Adaptation-of-Agentic-AI

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨TabReX : Tabular Referenceless eXplainable Evaluation

📝 Summary:
TabReX is a reference-less framework using graph-based reasoning to evaluate the quality of tables generated by LLMs, offering structural and factual fidelity scores. AI-generated summary Evaluating t...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15907
• PDF: https://arxiv.org/pdf/2512.15907
• Project Page: https://coral-lab-asu.github.io/TabReX/
• Github: https://github.com/CoRAL-ASU/TabReX

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Seedance 1.5 pro: A Native Audio-Visual Joint Generation Foundation Model

📝 Summary:
Seedance 1.5 pro, a dual-branch Diffusion Transformer model, achieves high-quality audio-visual synchronization and generation through cross-modal integration, post-training optimizations, and an acce...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13507
• PDF: https://arxiv.org/pdf/2512.13507
• Project Page: https://seed.bytedance.com/seedance1_5_pro

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Depth Any Panoramas: A Foundation Model for Panoramic Depth Estimation

📝 Summary:
A panoramic metric depth foundation model using DINOv3-Large and a three-stage pseudo-label pipeline achieves robust performance across diverse real-world scenes. AI-generated summary In this work, we...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16913
• PDF: https://arxiv.org/pdf/2512.16913
• Github: https://insta360-research-team.github.io/DAP

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Next-Embedding Prediction Makes Strong Vision Learners

📝 Summary:
Generative pretraining using next embedding prediction outperforms traditional self-supervised methods in visual learning tasks, achieving high accuracy on ImageNet and effective transfer to semantic ...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16922
• PDF: https://arxiv.org/pdf/2512.16922

🔹 Models citing this paper:
• https://huggingface.co/SixAILab/nepa-base-patch14-224-sft
• https://huggingface.co/SixAILab/nepa-large-patch14-224
• https://huggingface.co/SixAILab/nepa-base-patch14-224

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Alchemist: Unlocking Efficiency in Text-to-Image Model Training via Meta-Gradient Data Selection

📝 Summary:
Alchemist, a meta-gradient-based framework, automatically selects high-quality subsets from large-scale text-image datasets to improve visual quality and training efficiency in Text-to-Image models. A...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16905
• PDF: https://arxiv.org/pdf/2512.16905

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨StereoPilot: Learning Unified and Efficient Stereo Conversion via Generative Priors

📝 Summary:
StereoPilot, a feed-forward model leveraging a learnable domain switcher and cycle consistency loss, synthesizes high-quality stereo video directly without depth maps, outperforming existing methods i...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16915
• PDF: https://arxiv.org/pdf/2512.16915

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Exploration v.s. Exploitation: Rethinking RLVR through Clipping, Entropy, and Spurious Reward

📝 Summary:
Reinforcement learning with verifiable rewards improves LLM reasoning through spurious rewards and entropy minimization, despite seemingly paradoxical effects, by reducing clipping bias and policy ent...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16912
• PDF: https://arxiv.org/pdf/2512.16912

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Differences That Matter: Auditing Models for Capability Gap Discovery and Rectification

📝 Summary:
AuditDM, an automated framework using reinforcement learning, identifies and rectifies failure modes in multimodal LLMs by generating challenging examples, leading to improved performance across bench...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16921
• PDF: https://arxiv.org/pdf/2512.16921
• Project Page: https://auditdm.github.io/
• Github: https://auditdm.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EmoCaliber: Advancing Reliable Visual Emotion Comprehension via Confidence Verbalization and Calibration

📝 Summary:
EmoCaliber, a confidence-aware Multimodal Large Language Model, enhances Visual Emotion Comprehension by verbalizing confidence in emotion predictions, leading to improved reliability and accuracy. AI...

🔹 Publication Date: Published on Dec 17

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.15528
• PDF: https://arxiv.org/pdf/2512.15528
• Github: https://github.com/wdqqdw/EmoCaliber

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Generative Refocusing: Flexible Defocus Control from a Single Image

📝 Summary:
Generative Refocusing uses DeblurNet and BokehNet for high-quality single-image refocusing. Its semi-supervised training with real bokeh images and EXIF metadata enables controllable bokeh and text-guided adjustments, outperforming current methods.

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16923
• PDF: https://arxiv.org/pdf/2512.16923
• Project Page: https://generative-refocusing.github.io/
• Github: https://github.com/rayray9999/Genfocus

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RePlan: Reasoning-guided Region Planning for Complex Instruction-based Image Editing

📝 Summary:
RePlan, a plan-then-execute framework, enhances instruction-based image editing by combining a vision-language planner with a diffusion editor, achieving superior performance in complex and intricate ...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16864
• PDF: https://arxiv.org/pdf/2512.16864
• Project Page: https://replan-iv-edit.github.io/
• Github: https://github.com/dvlab-research/RePlan

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AdaTooler-V: Adaptive Tool-Use for Images and Videos

📝 Summary:
AdaTooler-V, a multimodal large language model, adaptively uses vision tools based on reinforcement learning, improving performance and reducing unnecessary tool invocations in visual reasoning tasks....

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16918
• PDF: https://arxiv.org/pdf/2512.16918
• Github: https://github.com/CYWang735/AdaTooler-V

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨N3D-VLM: Native 3D Grounding Enables Accurate Spatial Reasoning in Vision-Language Models

📝 Summary:
N3D-VLM integrates native 3D perception and reasoning in vision-language models, enabling precise 3D localization and spatial understanding with a large-scale dataset. AI-generated summary While curre...

🔹 Publication Date: Published on Dec 18

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.16561
• PDF: https://arxiv.org/pdf/2512.16561
• Github: https://github.com/W-Ted/N3D-VLM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research