✨Causal Judge Evaluation: Calibrated Surrogate Metrics for LLM Systems

📝 Summary:
CJE improves LLM-as-judge evaluation by fixing statistical issues like uncalibrated scores and poor confidence intervals. It achieves 99% ranking accuracy at 14x lower cost by calibrating a cheaper judge with 5% oracle labels.

🔹 Publication Date: Published on Dec 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.11150
• PDF: https://arxiv.org/pdf/2512.11150
• Project Page: https://www.cimolabs.com/cje
• Github: https://github.com/cimo-labs/cje

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

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

#LLMs #AIEvaluation #MachineLearning #DataScience #NLP

✨Particulate: Feed-Forward 3D Object Articulation

📝 Summary:
Particulate is a feed-forward method using a transformer network to infer articulated 3D structures from single static meshes, achieving faster and more accurate results than prior approaches. AI-gene...

🔹 Publication Date: Published on Dec 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.11798
• PDF: https://arxiv.org/pdf/2512.11798

🔹 Models citing this paper:
• https://huggingface.co/rayli/Particulate

✨ Spaces citing this paper:
• https://huggingface.co/spaces/rayli/particulate

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Fast-FoundationStereo: Real-Time Zero-Shot Stereo Matching

📝 Summary:
Fast-FoundationStereo enables real-time zero-shot stereo matching. It uses knowledge distillation, neural architecture search, and structured pruning to achieve this. The model runs over 10x faster than previous models while maintaining accuracy, setting a new state-of-the-art.

🔹 Publication Date: Published on Dec 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.11130
• PDF: https://arxiv.org/pdf/2512.11130
• Project Page: https://nvlabs.github.io/Fast-FoundationStereo/
• Github: https://github.com/NVlabs/Fast-FoundationStereo

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Interpretable Embeddings with Sparse Autoencoders: A Data Analysis Toolkit

📝 Summary:
Sparse autoencoders SAEs create interpretable, cost-effective embeddings where dimensions map to concepts. These SAE embeddings outperform LLMs and dense embeddings for large-scale text analysis, offering better control for tasks like bias identification and dataset comparison.

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.10092
• PDF: https://arxiv.org/pdf/2512.10092
• Project Page: https://interp-embed.com
• Github: https://github.com/nickjiang2378/interp_embed

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Openpi Comet: Competition Solution For 2025 BEHAVIOR Challenge

📝 Summary:
The Openpi Comet solution for the 2025 BEHAVIOR Challenge addresses household tasks using pre-training and post-training. It achieved a close second place, significantly outperforming other submissions, demonstrating the scaling power of these methods.

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2512.10071
• PDF: https://arxiv.org/pdf/2512.10071
• Github: https://github.com/mli0603/openpi-comet

🔹 Models citing this paper:
• https://huggingface.co/sunshk/comet_submission

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MentraSuite: Post-Training Large Language Models for Mental Health Reasoning and Assessment

📝 Summary:
MentraSuite, a unified framework, advances reliable mental health reasoning using Mindora, a post-trained model with hybrid SFT-RL, evaluated via MentraBench, a benchmark assessing task performance an...

🔹 Publication Date: Published on Dec 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.09636
• PDF: https://arxiv.org/pdf/2512.09636
• Github: https://github.com/elsa66666/MentraSuite

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Error-Free Linear Attention is a Free Lunch: Exact Solution from Continuous-Time Dynamics

📝 Summary:
Error-Free Linear Attention (EFLA) is a stable, parallelizable, and theoretically sound linear-time attention mechanism that outperforms DeltaNet in language modeling and downstream tasks. AI-generate...

🔹 Publication Date: Published on Dec 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.12602
• PDF: https://arxiv.org/pdf/2512.12602
• Github: https://github.com/declare-lab/EFLA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents

📝 Summary:
NL2Repo Bench evaluates long-horizon software development capabilities of coding agents by assessing their ability to generate complete Python libraries from natural-language requirements. AI-generate...

🔹 Publication Date: Published on Dec 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.12730
• PDF: https://arxiv.org/pdf/2512.12730
• Project Page: https://github.com/multimodal-art-projection/NL2RepoBench
• Github: https://github.com/multimodal-art-projection/NL2RepoBench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨LongVie 2: Multimodal Controllable Ultra-Long Video World Model

📝 Summary:
LongVie 2, an end-to-end autoregressive framework, enhances controllability, visual quality, and temporal consistency in video world models through three progressive training stages. AI-generated summ...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13604
• PDF: https://arxiv.org/pdf/2512.13604
• Project Page: https://vchitect.github.io/LongVie2-project/
• Github: https://github.com/Vchitect/LongVie

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨V-REX: Benchmarking Exploratory Visual Reasoning via Chain-of-Questions

📝 Summary:
The V-REX evaluation suite assesses vision-language models' multi-step reasoning and exploration capabilities through a Chain-of-Questions framework, revealing their strengths and weaknesses in planni...

🔹 Publication Date: Published on Dec 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.11995
• PDF: https://arxiv.org/pdf/2512.11995
• Github: https://github.com/tianyi-lab/VREX

✨ Datasets citing this paper:
• https://huggingface.co/datasets/umd-zhou-lab/V-REX

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Image Diffusion Preview with Consistency Solver

📝 Summary:
Diffusion Preview uses ConsistencySolver, a high-order trainable solver, to improve quality and consistency in low-step image generation, enhancing interactive user experiences. AI-generated summary T...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13592
• PDF: https://arxiv.org/pdf/2512.13592
• Github: https://github.com/G-U-N/consolver

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Video Reality Test: Can AI-Generated ASMR Videos fool VLMs and Humans?

📝 Summary:
The Video Reality Test benchmark evaluates the realism and detection of AI-generated ASMR videos with audio, revealing that even the best models can deceive VLMs and humans, highlighting limitations i...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13281
• PDF: https://arxiv.org/pdf/2512.13281
• Project Page: https://video-reality-test.github.io/
• Github: https://github.com/video-reality-test/video-reality-test

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Aesthetic Alignment Risks Assimilation: How Image Generation and Reward Models Reinforce Beauty Bias and Ideological "Censorship"

📝 Summary:
State-of-the-art image generation and reward models exhibit bias towards conventional aesthetics, often failing to produce anti-aesthetic images as requested, thus compromising user autonomy and aesth...

🔹 Publication Date: Published on Dec 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.11883
• PDF: https://arxiv.org/pdf/2512.11883

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨KlingAvatar 2.0 Technical Report

📝 Summary:
KlingAvatar 2.0 addresses inefficiencies in generating long-duration, high-resolution videos by using a spatio-temporal cascade framework with a Co-Reasoning Director and Negative Director for improve...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13313
• PDF: https://arxiv.org/pdf/2512.13313
• Project Page: https://app.klingai.com/global/ai-human/image/new/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management

📝 Summary:
QwenLong-L1.5 enhances long-context reasoning through data synthesis, stabilized reinforcement learning, and memory-augmented architecture, achieving superior performance on benchmarks and general dom...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.12967
• PDF: https://arxiv.org/pdf/2512.12967

🔹 Models citing this paper:
• https://huggingface.co/Tongyi-Zhiwen/QwenLong-L1.5-30B-A3B

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨START: Spatial and Textual Learning for Chart Understanding

📝 Summary:
START enhances multimodal large language models by integrating spatial and textual learning through chart-element grounding and chart-to-code generation, improving chart understanding and performance ...

🔹 Publication Date: Published on Dec 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.07186
• PDF: https://arxiv.org/pdf/2512.07186
• Github: https://github.com/dragonlzm/START

🔹 Models citing this paper:
• https://huggingface.co/zhuomingliu/START

✨ Datasets citing this paper:
• https://huggingface.co/datasets/zhuomingliu/CS-Bench
• https://huggingface.co/datasets/zhuomingliu/START-Dataset
• https://huggingface.co/datasets/zhuomingliu/START_eval

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Memory in the Age of AI Agents

📝 Summary:
This survey provides an updated overview of agent memory research, distinguishing its forms, functions, and dynamics, and highlights emerging research directions. AI-generated summary Memory has emerg...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13564
• PDF: https://arxiv.org/pdf/2512.13564

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ReFusion: A Diffusion Large Language Model with Parallel Autoregressive Decoding

📝 Summary:
ReFusion, a novel masked diffusion model, improves performance and efficiency by using slot-based parallel decoding, achieving superior results compared to autoregressive models and traditional masked...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13586
• PDF: https://arxiv.org/pdf/2512.13586
• Github: https://github.com/ML-GSAI/ReFusion

🔹 Models citing this paper:
• https://huggingface.co/GSAI-ML/ReFusion

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Spatial-Aware VLA Pretraining through Visual-Physical Alignment from Human Videos

📝 Summary:
A Spatial-Aware VLA Pretraining paradigm improves 3D spatial understanding in robots by aligning 2D visual inputs with 3D actions using dual-encoder architecture with a 3D visual encoder. AI-generated...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13080
• PDF: https://arxiv.org/pdf/2512.13080
• Project Page: https://beingbeyond.github.io/VIPA-VLA/
• Github: https://beingbeyond.github.io/VIPA-VLA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Toward Ambulatory Vision: Learning Visually-Grounded Active View Selection

📝 Summary:
VG-AVS, a task and framework fine-tunes VLMs to select the most informative next viewpoint for visual question answering, enhancing performance and generalization. AI-generated summary Vision Language...

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13250
• PDF: https://arxiv.org/pdf/2512.13250
• Project Page: https://active-view-selection.github.io
• Github: https://github.com/KAIST-Visual-AI-Group/VG-AVS

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨LitePT: Lighter Yet Stronger Point Transformer

📝 Summary:
LitePT combines early convolutions and deep attention for 3D point clouds, using PointROPE positional encoding. This new model is highly efficient, outperforming state-of-the-art while using fewer resources.

🔹 Publication Date: Published on Dec 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.13689
• PDF: https://arxiv.org/pdf/2512.13689
• Project Page: https://litept.github.io/
• Github: https://github.com/prs-eth/LitePT

🔹 Models citing this paper:
• https://huggingface.co/yuanwenyue/LitePT

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research