✨MSA: Memory Sparse Attention for Efficient End-to-End Memory Model Scaling to 100M Tokens

📝 Summary:
Memory Sparse Attention (MSA) enables large language models to process extremely long contexts with linear complexity and high efficiency through innovations like sparse attention and document-wise Ro...

🔹 Publication Date: Published on Mar 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.23516
• PDF: https://arxiv.org/pdf/2603.23516
• Project Page: https://evermind.ai/blogs/breaking-the-100m-token-limit-msa-architecture-achieves-efficient-end-to-end-long-term-memory-for-llms
• Github: https://github.com/EverMind-AI/MSA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Voxtral TTS

📝 Summary:
Voxtral TTS is a multilingual text-to-speech model that generates natural speech from short reference audio using a hybrid architecture combining semantic token generation and flow-matching for acoust...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25551
• PDF: https://arxiv.org/pdf/2603.25551
• Project Page: https://mistral.ai/news/voxtral-tts

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting

📝 Summary:
LGTM is a feed-forward framework that enables high-fidelity 4K novel view synthesis by predicting compact Gaussian primitives with per-primitive textures, decoupling geometric complexity from renderin...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25745
• PDF: https://arxiv.org/pdf/2603.25745
• Project Page: https://yxlao.github.io/lgtm/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Less Gaussians, Texture More: 4K Feed-Forward Textured Splatting

📝 Summary:
LGTM is a feed-forward framework that enables high-fidelity 4K novel view synthesis by predicting compact Gaussian primitives with per-primitive textures, decoupling geometric complexity from renderin...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25745
• PDF: https://arxiv.org/pdf/2603.25745
• Project Page: https://yxlao.github.io/lgtm/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Vega: Learning to Drive with Natural Language Instructions

📝 Summary:
Vega is a unified Vision-Language-World-Action model that combines autoregressive and diffusion paradigms for instruction-based driving planning and trajectory generation. AI-generated summary Vision-...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25741
• PDF: https://arxiv.org/pdf/2603.25741
• Project Page: https://zuosc19.github.io/Vega/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale

📝 Summary:
Intern-S1-Pro is a one-trillion-parameter scientific multimodal foundation model that enhances general and scientific capabilities through advanced agent functionalities and specialized task mastery a...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25040
• PDF: https://arxiv.org/pdf/2603.25040

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MuRF: Unlocking the Multi-Scale Potential of Vision Foundation Models

📝 Summary:
Multi-Resolution Fusion enables vision foundation models to leverage complementary inductive biases from different resolutions without architectural modifications or additional training. AI-generated ...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25744
• PDF: https://arxiv.org/pdf/2603.25744
• Project Page: https://MuRF-VFM.github.io
• Github: https://github.com/orgs/MuRF-VFM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨BioVITA: Biological Dataset, Model, and Benchmark for Visual-Textual-Acoustic Alignment

📝 Summary:
A multimodal framework for biological species identification that aligns visual, textual, and acoustic data to learn unified representations capturing species-level semantics beyond traditional taxono...

🔹 Publication Date: Published on Mar 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.23883
• PDF: https://arxiv.org/pdf/2603.23883
• Project Page: https://dahlian00.github.io/BioVITA_Page/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨FinMCP-Bench: Benchmarking LLM Agents for Real-World Financial Tool Use under the Model Context Protocol

📝 Summary:
FinMCP-Bench is a comprehensive benchmark for evaluating large language models on financial problem-solving through tool invocation and reasoning across multiple complexity levels. AI-generated summar...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.24943
• PDF: https://arxiv.org/pdf/2603.24943
• Project Page: https://github.com/aliyun/qwen-dianjin

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Extending Precipitation Nowcasting Horizons via Spectral Fusion of Radar Observations and Foundation Model Priors

📝 Summary:
Deep learning model for precipitation nowcasting that combines radar imagery with meteorological forecasts through frequency-domain fusion techniques to improve long-term forecasting accuracy. AI-gene...

🔹 Publication Date: Published on Mar 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.21768
• PDF: https://arxiv.org/pdf/2603.21768
• Github: https://github.com/Onemissed/PW-FouCast

🔹 Models citing this paper:
• https://huggingface.co/Onemiss/PW-FouCast

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨PixelSmile: Toward Fine-Grained Facial Expression Editing

📝 Summary:
PixelSmile is a diffusion framework for fine-grained facial expression editing. It achieves better disentanglement and identity preservation through symmetric joint training and contrastive learning. This enables precise, stable, and continuous control for expression editing.

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25728
• PDF: https://arxiv.org/pdf/2603.25728
• Project Page: https://ammmob.github.io/PixelSmile/
• Github: https://github.com/Ammmob/PixelSmile

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

✨ Datasets citing this paper:
• https://huggingface.co/datasets/PixelSmile/FFE-Bench

✨ Spaces citing this paper:
• https://huggingface.co/spaces/Pr0f3ssi0n4ln00b/Qwen-Image-Edit-Rapid-AIO-Loras-Experimental
• https://huggingface.co/spaces/PixelSmile/PixelSmile-Demo

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

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

#FacialExpressionEditing #DiffusionModels #AI #ComputerVision #DeepLearning

✨RealRestorer: Towards Generalizable Real-World Image Restoration with Large-Scale Image Editing Models

📝 Summary:
A large-scale dataset and open-source model are developed to improve image restoration performance and close the gap with closed-source alternatives, with a dedicated benchmark for real-world degradat...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/pdf/2603.25502
• PDF: https://arxiv.org/pdf/2603.25502
• Project Page: https://yfyang007.github.io/RealRestorer/
• Github: https://github.com/yfyang007/RealRestorer

🔹 Models citing this paper:
• https://huggingface.co/RealRestorer/RealRestorer
• https://huggingface.co/RealRestorer/RealRestorer_degradation_models

✨ Datasets citing this paper:
• https://huggingface.co/datasets/RealRestorer/RealIR-Bench

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Representation Alignment for Just Image Transformers is not Easier than You Think

📝 Summary:
Representation alignment fails for pixel-space diffusion transformers due to information asymmetry, but PixelREPA addresses this by transforming alignment targets and using masked transformer adapters...

🔹 Publication Date: Published on Mar 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.14366
• PDF: https://arxiv.org/pdf/2603.14366
• Github: https://github.com/kaist-cvml/PixelREPA

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨S2D2: Fast Decoding for Diffusion LLMs via Training-Free Self-Speculation

📝 Summary:
S2D2 is a training-free self-speculative decoding framework for block-diffusion LLMs. It improves accuracy-speed by using the same model as both parallel drafter and autoregressive verifier via a speculative verification step. This achieves significant speedups up to 4.7 times and higher accuracy...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25702
• PDF: https://arxiv.org/pdf/2603.25702
• Github: https://github.com/phymhan/S2D2

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

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

#LLM #DiffusionModels #Decoding #AI #MachineLearning

✨Electrostatic Photoluminescence Tuning in All-Solid-State Perovskite Transistors

📝 Summary:
This paper demonstrates an all-solid-state perovskite transistor that electrostatically controls photoluminescence intensity. By modulating charge recombination, it achieves high quantum efficiencies and tunable light emission. This expands perovskite applications in photonics.

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25718
• PDF: https://arxiv.org/pdf/2603.25718
• Project Page: https://kj-chen666.github.io/Hybrid-Memory-in-Video-World-Models/
• Github: https://github.com/H-EmbodVis/HyDRA

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

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

#Perovskites #Photoluminescence #Optoelectronics #Transistors #Photonics

✨Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes

📝 Summary:
On-policy distillation for LLMs suffers from fragile token-level signals and unreliable teacher guidance. This paper introduces teacher top-K local support matching with truncated reverse-KL, top-p sampling, and special-token masking to achieve stable optimization and improved performance.

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25562
• PDF: https://arxiv.org/pdf/2603.25562
• Project Page: https://www.notion.so/yuqianfu/Revisiting-On-Policy-Distillation-Empirical-Failure-Modes-and-Simple-Fixes-31dd5cc40dd181f89eead3de7181df1d
• Github: https://github.com/hhh675597/revisiting_opd

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

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

#OnPolicyDistillation #LLMs #MachineLearning #DeepLearning #NLP

✨MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-Evolution

📝 Summary:
MemMA is a multi-agent framework that coordinates the memory cycle in LLM agents. It uses a Meta-Thinker for strategic guidance and in-situ self-evolving repair for memory construction and retrieval. MemMA consistently outperforms existing baselines.

🔹 Publication Date: Published on Mar 19

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.18718
• PDF: https://arxiv.org/pdf/2603.18718
• Github: https://github.com/ventr1c/memma

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

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

#LLM #MultiAgentSystems #AIMemory #AIResearch #ArtificialIntelligence

✨Calibri: Enhancing Diffusion Transformers via Parameter-Efficient Calibration

📝 Summary:
Calibri enhances Diffusion Transformers by adding a single learned scaling parameter to improve generative quality. This parameter-efficient method, optimizing only ~100 parameters, reduces inference steps across various text-to-image models while maintaining high-quality outputs.

🔹 Publication Date: Published on Mar 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.24800
• PDF: https://arxiv.org/pdf/2603.24800
• Project Page: https://v-gen-ai.github.io/Calibri-page/
• Github: https://github.com/v-gen-ai/Calibri

🔹 Models citing this paper:
• https://huggingface.co/v-gen-ai/flux-calibri-gates
• https://huggingface.co/v-gen-ai/qwen-calibri

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

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

#DiffusionModels #GenerativeAI #AIResearch #MachineLearning #DeepLearning

✨AVControl: Efficient Framework for Training Audio-Visual Controls

📝 Summary:
AVControl efficiently enables modular audio-visual generation by training diverse controls as separate LoRA adapters on a parallel canvas in LTX-2. It achieves superior performance on various tasks including depth and pose guidance, requiring minimal computational resources.

🔹 Publication Date: Published on Mar 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.24793
• PDF: https://arxiv.org/pdf/2603.24793
• Project Page: https://matanby.github.io/AVControl/

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

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

#AudioVisualAI #GenerativeAI #LoRA #EfficientAI #DeepLearning

✨PMT: Plain Mask Transformer for Image and Video Segmentation with Frozen Vision Encoders

📝 Summary:
PMT introduces a Plain Mask Decoder for fast image and video segmentation using frozen Vision Foundation Model encoders. This preserves VFM multi-task sharing, achieving competitive accuracy and significant speed improvements over prior methods.

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25398
• PDF: https://arxiv.org/pdf/2603.25398
• Github: https://github.com/tue-mps/pmt

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

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

#ImageSegmentation #VideoSegmentation #Transformers #ComputerVision #DeepLearning