✨ROI-Reasoning: Rational Optimization for Inference via Pre-Computation Meta-Cognition

📝 Summary:
ROI Reasoning enables large language models to strategically allocate computation under strict token budgets. It uses meta-cognition to predict costs and utilities, optimizing sequential decisions with reinforcement learning. This improves performance and reduces regret on budgeted reasoning tasks.

🔹 Publication Date: Published on Jan 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03822
• PDF: https://arxiv.org/pdf/2601.03822

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RelayLLM: Efficient Reasoning via Collaborative Decoding

📝 Summary:
RelayLLM enables efficient collaborative reasoning by having a small language model dynamically invoke a large language model only for critical tokens. This token-level collaboration achieves high accuracy with minimal computational overhead. It reduces LLM invocation to just 1.07% of tokens, lea...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05167
• PDF: https://arxiv.org/pdf/2601.05167

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twice

📝 Summary:
VideoAuto-R1 framework employs a reason-when-necessary strategy for video understanding, using a Thinking Once, Answering Twice training paradigm with verifiable rewards and confidence-based reasoning...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05175
• PDF: https://arxiv.org/pdf/2601.05175
• Project Page: https://ivul-kaust.github.io/projects/videoauto-r1/

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨VerseCrafter: Dynamic Realistic Video World Model with 4D Geometric Control

📝 Summary:
VerseCrafter is a 4D-aware video world model that enables unified control over camera and object dynamics through 4D geometric control representation and video diffusion models. AI-generated summary V...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05138
• PDF: https://arxiv.org/pdf/2601.05138

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Agent-as-a-Judge

📝 Summary:
Large language models face limitations in evaluating complex, multi-step tasks, prompting the development of agent-based evaluation systems that utilize planning, tool-augmented verification, and mult...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05111
• PDF: https://arxiv.org/pdf/2601.05111

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DiffCoT: Diffusion-styled Chain-of-Thought Reasoning in LLMs

📝 Summary:
DiffCoT reformulates chain-of-thought reasoning as an iterative denoising process using diffusion principles, enabling unified generation and correction of intermediate steps while maintaining causal ...

🔹 Publication Date: Published on Jan 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03559
• PDF: https://arxiv.org/pdf/2601.03559

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨The Illusion of Specialization: Unveiling the Domain-Invariant "Standing Committee" in Mixture-of-Experts Models

📝 Summary:
Research challenges the assumption of domain specialization in Mixture of Experts models by identifying a persistent central committee of experts that dominates routing behavior across different domai...

🔹 Publication Date: Published on Jan 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03425
• PDF: https://arxiv.org/pdf/2601.03425

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization

📝 Summary:
Multi-reward RL with GRPO suffers from reward normalization collapse, leading to suboptimal training. GDPO solves this by decoupling individual reward normalization, preserving their relative differences for improved stability and optimization. GDPO consistently outperforms GRPO across various re...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05242
• PDF: https://arxiv.org/pdf/2601.05242
• Project Page: https://nvlabs.github.io/GDPO/
• Github: https://github.com/NVlabs/GDPO

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Plenoptic Video Generation

📝 Summary:
PlenopticDreamer enables consistent multi-view video re-rendering through synchronized generative hallucinations, leveraging camera-guided retrieval and progressive training mechanisms for improved te...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05239
• PDF: https://arxiv.org/pdf/2601.05239
• Project Page: https://research.nvidia.com/labs/dir/plenopticdreamer/

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DocDancer: Towards Agentic Document-Grounded Information Seeking

📝 Summary:
DocDancer is an end-to-end trained open-source document question answering agent that formulates the task as an information-seeking problem and uses a tool-driven framework with exploration and synthe...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05163
• PDF: https://arxiv.org/pdf/2601.05163

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Token-Level LLM Collaboration via FusionRoute

📝 Summary:
FusionRoute is a token-level multi-LLM collaboration framework that uses a lightweight router to select optimal experts and add complementary logits, outperforming existing methods in diverse tasks wh...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05106
• PDF: https://arxiv.org/pdf/2601.05106

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Safety at One Shot: Patching Fine-Tuned LLMs with A Single Instance

📝 Summary:
Safety alignment of large language models can be fully recovered with a single safety example, maintaining utility and achieving convergence in few epochs through identified low-rank gradient structur...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.01887
• PDF: https://arxiv.org/pdf/2601.01887

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Memorization in 3D Shape Generation: An Empirical Study

📝 Summary:
Researchers develop a framework to measure memorization in 3D generative models and identify factors affecting it, finding that data modality and model design parameters influence how much training da...

🔹 Publication Date: Published on Dec 29, 2025

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.23628
• PDF: https://arxiv.org/pdf/2512.23628
• Github: https://github.com/zlab-princeton/3d-gen-mem

🔹 Models citing this paper:
• https://huggingface.co/pudashi/3DGenMem

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RoboVIP: Multi-View Video Generation with Visual Identity Prompting Augments Robot Manipulation

📝 Summary:
Visual identity prompting enhances manipulation data augmentation for robot policies by providing explicit visual guidance to diffusion models, improving policy performance in both simulation and real...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05241
• PDF: https://arxiv.org/pdf/2601.05241
• Project Page: https://robovip.github.io/RoboVIP/
• Github: https://robovip.github.io/RoboVIP/

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Learnable Multipliers: Freeing the Scale of Language Model Matrix Layers

📝 Summary:
Learnable multipliers are introduced to address weight decay-induced normalization artifacts in large language model training, outperforming traditional methods while reducing computational overhead. ...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04890
• PDF: https://arxiv.org/pdf/2601.04890
• Project Page: https://tiiuae.github.io/Falcon-H1/
• Github: https://github.com/tiiuae/falcon-h1

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Re-Align: Structured Reasoning-guided Alignment for In-Context Image Generation and Editing

📝 Summary:
Re-Align addresses the gap between understanding and generation in in-context image generation and editing through structured reasoning-guided alignment and reinforcement learning training. AI-generat...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05124
• PDF: https://arxiv.org/pdf/2601.05124
• Project Page: https://hrz2000.github.io/realign/
• Github: https://github.com/hrz2000/realign

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Guardians of the Hair: Rescuing Soft Boundaries in Depth, Stereo, and Novel Views

📝 Summary:
HairGuard is a framework designed to recover fine-grained soft boundary details in 3D vision tasks. It refines depth around these ambiguous regions and synthesizes novel views, achieving state-of-the-art performance for delicate structures like hair.

🔹 Publication Date: Published on Jan 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.03362
• PDF: https://arxiv.org/pdf/2601.03362

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Enhancing Object Detection with Privileged Information: A Model-Agnostic Teacher-Student Approach

📝 Summary:
Learning Using Privileged Information paradigm enhances object detection accuracy by integrating additional training-time information through teacher-student architectures without increasing inference...

🔹 Publication Date: Published on Jan 5

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.02016
• PDF: https://arxiv.org/pdf/2601.02016

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨AT^2PO: Agentic Turn-based Policy Optimization via Tree Search

📝 Summary:
AT²PO is a unified framework for multi-turn agentic reinforcement learning that improves exploration diversity, credit assignment, and policy optimization through tree search and turn-level learning o...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04767
• PDF: https://arxiv.org/pdf/2601.04767
• Github: https://github.com/zzfoutofspace/ATPO

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

For more data science resources:
✓ https://t.me/DataScienceT

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RL-AWB: Deep Reinforcement Learning for Auto White Balance Correction in Low-Light Night-time Scenes

📝 Summary:
RL-AWB is a novel framework combining statistical methods with deep reinforcement learning for improved nighttime auto white balance. It is the first RL approach for color constancy, mimicking expert tuning. This method shows superior generalization across various lighting conditions, and a new m...

🔹 Publication Date: Published on Jan 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.05249
• PDF: https://arxiv.org/pdf/2601.05249
• Project Page: https://ntuneillee.github.io/research/rl-awb/

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

For more data science resources:
✓ https://t.me/DataScienceT

#ReinforcementLearning #ComputerVision #ImageProcessing #AutoWhiteBalance #LowLightImaging

✨Beyond Binary Preference: Aligning Diffusion Models to Fine-grained Criteria by Decoupling Attributes

📝 Summary:
Current diffusion model alignment struggles with complex, fine-grained human expertise due to simplified preferences. This paper proposes a framework with hierarchical criteria and Complex Preference Optimization CPO, maximizing positive and minimizing negative attributes to improve generation qu...

🔹 Publication Date: Published on Jan 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2601.04300
• PDF: https://arxiv.org/pdf/2601.04300

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

For more data science resources:
✓ https://t.me/DataScienceT

#DiffusionModels #AIAlignment #MachineLearning #GenerativeAI #PreferenceLearning