✨SoundWeaver: Semantic Warm-Starting for Text-to-Audio Diffusion Serving

📝 Summary:
SoundWeaver accelerates text-to-audio diffusion generation by caching semantically similar audio and dynamically skipping function evaluations, achieving significant latency reduction with minimal qua...

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.07865
• PDF: https://arxiv.org/pdf/2603.07865

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DreamVideo-Omni: Omni-Motion Controlled Multi-Subject Video Customization with Latent Identity Reinforcement Learning

📝 Summary:
DreamVideo-Omni is a unified framework for video synthesis that enables precise multi-subject identity control and multi-granularity motion manipulation through a two-stage training approach combining...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12257
• PDF: https://arxiv.org/pdf/2603.12257
• Project Page: https://dreamvideo-omni.github.io/
• Github: https://dreamvideo-omni.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training

📝 Summary:
Research examines the effectiveness of reasoning versus non-reasoning large language model judges in reinforcement learning-based alignment, revealing that reasoning judges prevent reward hacking but ...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12246
• PDF: https://arxiv.org/pdf/2603.12246

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨One Model, Many Budgets: Elastic Latent Interfaces for Diffusion Transformers

📝 Summary:
Elastic Latent Interface Transformer (ELIT) decouples compute from image resolution in diffusion transformers by introducing learnable latent tokens that adaptively prioritize important regions, enabl...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12245
• PDF: https://arxiv.org/pdf/2603.12245
• Project Page: https://snap-research.github.io/elit/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Multi-Task Reinforcement Learning for Enhanced Multimodal LLM-as-a-Judge

📝 Summary:
Multi-Task Reinforcement Learning framework improves multimodal large language models' judgment consistency and generalization across diverse visual tasks. AI-generated summary Multimodal Large Langua...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11665
• PDF: https://arxiv.org/pdf/2603.11665

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks

📝 Summary:
Softmax self-attention models exhibit attention sinks where probability mass concentrates on fixed positions due to normalization constraints, while ReLU attention avoids this behavior. AI-generated s...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11487
• PDF: https://arxiv.org/pdf/2603.11487
• Github: https://github.com/YuvMilo/sinks-are-provably-necessary

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Understanding by Reconstruction: Reversing the Software Development Process for LLM Pretraining

📝 Summary:
Large language models trained on reconstructed agent trajectories from multi-agent simulations show improved performance in long-context understanding, coding proficiency, and agentic capabilities. AI...

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11103
• PDF: https://arxiv.org/pdf/2603.11103

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse

📝 Summary:
IndexCache reduces sparse attention computation in large language models by reusing top-k token selections across layers, achieving significant speedups with minimal quality loss. AI-generated summary...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12201
• PDF: https://arxiv.org/pdf/2603.12201

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EVATok: Adaptive Length Video Tokenization for Efficient Visual Autoregressive Generation

📝 Summary:
EVATok is a framework for efficient video tokenization that adapts token assignment based on video content, improving reconstruction quality and generation efficiency through learned routers and adapt...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12267
• PDF: https://arxiv.org/pdf/2603.12267
• Project Page: https://silentview.github.io/EVATok/
• Github: https://github.com/HKU-MMLab/EVATok

🔹 Models citing this paper:
• https://huggingface.co/YuuTennYi/EVATok

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DIVE: Scaling Diversity in Agentic Task Synthesis for Generalizable Tool Use

📝 Summary:
Training Qwen3-8B on DIVE data improves performance across out-of-distribution benchmarks, with diversity scaling outperforming quantity scaling even with less data. AI-generated summary Recent work s...

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11076
• PDF: https://arxiv.org/pdf/2603.11076
• Project Page: https://sheep333c.github.io/DIVE/
• Github: https://github.com/sheep333c/DIVE

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨EmbTracker: Traceable Black-box Watermarking for Federated Language Models

📝 Summary:
EmbTracker is a server-side black-box watermarking framework for federated language models that provides client-level traceability through unique identity-specific watermarks embedded via backdoor det...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12089
• PDF: https://arxiv.org/pdf/2603.12089

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Mobile-GS: Real-time Gaussian Splatting for Mobile Devices

📝 Summary:
Mobile-GS enables real-time 3D Gaussian Splatting rendering on mobile devices through depth-aware order-independent rendering, neural view-dependent enhancement, and compression techniques. AI-generat...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11531
• PDF: https://arxiv.org/pdf/2603.11531
• Project Page: https://xiaobiaodu.github.io/mobile-gs-project/
• Github: https://github.com/xiaobiaodu/mobile-gs

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DVD: Deterministic Video Depth Estimation with Generative Priors

📝 Summary:
DVD adapts pre-trained video diffusion models into deterministic single-pass depth regressors using structural anchors, latent manifold rectification, and global affine coherence. This framework achieves state-of-the-art zero-shot video depth estimation with significantly less data, overcoming li...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12250
• PDF: https://arxiv.org/pdf/2603.12250
• Project Page: https://dvd-project.github.io/
• Github: https://github.com/EnVision-Research/DVD

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Automatic Generation of High-Performance RL Environments

📝 Summary:
Automated framework generates high-performance reinforcement learning environments through prompt-based translation and verification, achieving significant speedups over existing implementations while...

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12145
• PDF: https://arxiv.org/pdf/2603.12145

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨FireRedASR2S: A State-of-the-Art Industrial-Grade All-in-One Automatic Speech Recognition System

📝 Summary:
FireRedASR2S is an industrial-grade ASR system integrating unified modules for speech recognition, voice activity detection, language identification, and punctuation prediction, achieving state-of-the...

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10420
• PDF: https://arxiv.org/pdf/2603.10420
• Project Page: https://github.com/FireRedTeam/FireRedASR2S
• Github: https://github.com/FireRedTeam/FireRedASR2S

🔹 Models citing this paper:
• https://huggingface.co/FireRedTeam/FireRedVAD
• https://huggingface.co/FireRedTeam/FireRedASR2-AED
• https://huggingface.co/FireRedTeam/FireRedASR2-LLM

✨ Spaces citing this paper:
• https://huggingface.co/spaces/FireRedTeam/FireRedASR2S

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Accent Vector: Controllable Accent Manipulation for Multilingual TTS Without Accented Data

📝 Summary:
Accent Vector enables controllable accent manipulation in multilingual TTS systems through fine-tuning on native speech from different languages and computing task vectors that capture accent characte...

🔹 Publication Date: Published on Mar 8

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.07534
• PDF: https://arxiv.org/pdf/2603.07534

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨The Curse and Blessing of Mean Bias in FP4-Quantized LLM Training

📝 Summary:
LLM anisotropy caused by rank-one mean bias in low-bit training can be stabilized through mean subtraction, recovering performance while enabling efficient hardware deployment. AI-generated summary La...

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10444
• PDF: https://arxiv.org/pdf/2603.10444

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨4DEquine: Disentangling Motion and Appearance for 4D Equine Reconstruction from Monocular Video

📝 Summary:
4DEquine is a new framework for 4D equine reconstruction from monocular video. It disentangles motion using spatio-temporal transformers and appearance with 3D Gaussian avatars. Training on synthetic data, it achieves state-of-the-art results on real-world datasets.

🔹 Publication Date: Published on Mar 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10125
• PDF: https://arxiv.org/pdf/2603.10125
• Project Page: https://luoxue-star.github.io/4DEquine_Project_Page/
• Github: https://github.com/luoxue-star/4DEquine

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

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

#ComputerVision #4DReconstruction #DeepLearning #Equine #AI

✨Training Language Models via Neural Cellular Automata

📝 Summary:
This paper introduces using Neural Cellular Automata NCA to generate synthetic data for pre-pre-training language models, addressing natural language limitations. This approach improves performance, accelerates convergence, and transfers to reasoning tasks, often outperforming extensive natural l...

🔹 Publication Date: Published on Mar 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.10055
• PDF: https://arxiv.org/pdf/2603.10055

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

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

#AI #LanguageModels #NeuralCellularAutomata #SyntheticData #NLP

✨XSkill: Continual Learning from Experience and Skills in Multimodal Agents

📝 Summary:
XSkill is a dual-stream framework for continual learning in multimodal agents. It extracts and retrieves knowledge from visual observations, consolidating experiences and skills. This improves tool use efficiency, reasoning, and zero-shot generalization.

🔹 Publication Date: Published on Mar 12

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.12056
• PDF: https://arxiv.org/pdf/2603.12056
• Project Page: https://xskill-agent.github.io/xskill_page/
• Github: https://github.com/XSkill-Agent/XSkill

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

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

#ContinualLearning #MultimodalAI #AIagents #MachineLearning #Robotics

✨Neural Field Thermal Tomography: A Differentiable Physics Framework for Non-Destructive Evaluation

📝 Summary:
NeFTY is a new differentiable physics framework that reconstructs 3D material properties from temperature measurements. It uses continuous neural fields and hard constraints, overcoming prior limitations and accurately locating subsurface defects.

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.11045
• PDF: https://arxiv.org/pdf/2603.11045
• Project Page: https://cab-lab-princeton.github.io/nefty/
• Github: https://cab-lab-princeton.github.io/nefty/

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

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

#NeuralFields #DifferentiablePhysics #NDE #MaterialScience #DeepLearning