✨TIP: Token Importance in On-Policy Distillation

📝 Summary:
On-policy knowledge distillation token selection methods are improved by identifying informative tokens through student entropy and teacher-student divergence, enabling efficient training with reduced...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.14084
• PDF: https://arxiv.org/pdf/2604.14084
• Github: https://github.com/HJSang/OPSD_OnPolicyDistillation

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨MERRIN: A Benchmark for Multimodal Evidence Retrieval and Reasoning in Noisy Web Environments

📝 Summary:
MERRIN is a human-annotated benchmark for evaluating search-augmented agents in multimodal, noisy web environments, demonstrating significant challenges in retrieving and reasoning over diverse eviden...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13418
• PDF: https://arxiv.org/pdf/2604.13418
• Project Page: https://merrin-benchmark.github.io
• Github: https://merrin-benchmark.github.io

✨ Datasets citing this paper:
• https://huggingface.co/datasets/HanNight/MERRIN

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨UI-Copilot: Advancing Long-Horizon GUI Automation via Tool-Integrated Policy Optimization

📝 Summary:
UI-Copilot is a collaborative framework that enhances GUI agents by decoupling memory management and integrating on-demand tool assistance for improved performance in complex user interface tasks. AI-...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13822
• PDF: https://arxiv.org/pdf/2604.13822
• Github: https://github.com/ZJU-REAL/UI-Copilot

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨RationalRewards: Reasoning Rewards Scale Visual Generation Both Training and Test Time

📝 Summary:
Training reward models to generate multi-dimensional critiques improves visual generation through both enhanced reinforcement learning rewards and test-time refinement loops, achieving state-of-the-ar...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11626
• PDF: https://arxiv.org/pdf/2604.11626
• Project Page: https://tiger-ai-lab.github.io/RationalRewards/
• Github: https://github.com/TIGER-AI-Lab/RationalRewards

🔹 Models citing this paper:
• https://huggingface.co/TIGER-Lab/RationalRewards-8B-T2I
• https://huggingface.co/TIGER-Lab/RationalRewards-8B-Edit

✨ Datasets citing this paper:
• https://huggingface.co/datasets/TIGER-Lab/RationalRewards_DiffusionNFT_TrainData

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Free Geometry: Refining 3D Reconstruction from Longer Versions of Itself

📝 Summary:
Free Geometry enables feed-forward 3D reconstruction models to self-evolve at test time through self-supervised cross-view feature consistency, improving reconstruction accuracy with lightweight LoRA ...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.14048
• PDF: https://arxiv.org/pdf/2604.14048
• Github: https://github.com/hiteacherIamhumble/Free-Geometry

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Space

📝 Summary:
PreRL applies reward-driven online updates to the marginal distribution in pre-train space, while DSRL uses NSR-PreRL to expand reasoning horizons before standard RL fine-tuning. AI-generated summary ...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.14142
• PDF: https://arxiv.org/pdf/2604.14142
• Github: https://github.com/Trae1ounG/Pretrain_Space_RLVR

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Exploration and Exploitation Errors Are Measurable for Language Model Agents

📝 Summary:
Controllable environments with programmable exploration-exploitation balance are designed to evaluate language model agents' performance on embodied AI tasks, revealing distinct failure modes and demo...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.13151
• PDF: https://arxiv.org/pdf/2604.13151
• Github: https://github.com/jjj-madison/measurable-explore-exploit

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Do AI Coding Agents Log Like Humans? An Empirical Study

📝 Summary:
S o f t w a r e l o g g i n g i s e s s e n t i a l f o r m a i n t a i n i n g a n d d e b u g g i n g c o m p l e x s y s t e m s , y e t i t r e m a i n s u n c l e a r h o w A I c o d i n g a g e ...

🔹 Publication Date: Published on Apr 10

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.09409
• PDF: https://arxiv.org/pdf/2604.09409

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Memory Transfer Learning: How Memories are Transferred Across Domains in Coding Agents

📝 Summary:
Memory Transfer Learning uses a unified memory pool from diverse coding domains to improve agent performance. It primarily transfers high-level meta-knowledge, not low-level code, showing that abstraction dictates effective cross-domain memory transfer.

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.14004
• PDF: https://arxiv.org/pdf/2604.14004
• Project Page: https://memorytransfer.github.io/
• Github: https://github.com/KangsanKim07/MemoryTransferLearning

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Sema Code: Decoupling AI Coding Agents into Programmable, Embeddable Infrastructure

📝 Summary:
Sema Code presents an open AI coding framework that decouples the core agent engine from client interfaces, enabling shared reasoning capabilities across diverse development environments through a sta...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11045
• PDF: https://arxiv.org/pdf/2604.11045
• Github: https://github.com/midea-ai/SemaClaw

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ReconPhys: Reconstruct Appearance and Physical Attributes from Single Video

📝 Summary:
ReconPhys is the first feedforward framework to jointly learn physical attribute estimation and 3D Gaussian Splatting reconstruction from a single video. It offers significantly faster inference and superior reconstruction quality for non-rigid objects compared to prior optimization-based methods...

🔹 Publication Date: Published on Apr 9

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.07882
• PDF: https://arxiv.org/pdf/2604.07882
• Project Page: https://chuanshuogushi.github.io/ReconPhys/
• Github: https://chuanshuogushi.github.io/ReconPhys/

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

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

#ComputerVision #3DReconstruction #GaussianSplatting #DeepLearning #AIResearch

✨SemaClaw: A Step Towards General-Purpose Personal AI Agents through Harness Engineering

📝 Summary:
SemaClaw is an open-source multi-agent framework addressing the need for robust infrastructure for personal AI agents. It ensures control and trustworthiness through novel orchestration, safety, and context management components, advancing general-purpose personal AI via harness engineering.

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11548
• PDF: https://arxiv.org/pdf/2604.11548
• Github: https://github.com/midea-ai/sema-code-core

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Mobile GUI Agents under Real-world Threats: Are We There Yet?

📝 Summary:
Mobile GUI agents powered by large language models show significant performance degradation when exposed to real-world third-party content in commercial applications. AI-generated summary Recent years...

🔹 Publication Date: Published on Apr 14

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2507.04227
• PDF: https://arxiv.org/pdf/2507.04227
• Project Page: https://agenthazard.github.io
• Github: https://github.com/Zsbyqx20/AgentHazard

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Target Policy Optimization

📝 Summary:
Target Policy Optimization separates policy update decisions from probability assignment in reinforcement learning, improving performance over standard policy gradient methods in sparse reward scenari...

🔹 Publication Date: Published on Apr 7

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.06159
• PDF: https://arxiv.org/pdf/2604.06159
• Github: https://github.com/JeanKaddour/tpo

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Feed-Forward 3D Scene Modeling: A Problem-Driven Perspective

📝 Summary:
This survey focuses on generalizable feed-forward 3D reconstruction, which efficiently maps images to 3D representations. It proposes a novel taxonomy centered on model design strategies, addressing key problems like feature enhancement and model efficiency, rather than output format differences.

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.14025
• PDF: https://arxiv.org/pdf/2604.14025
• Project Page: https://ff3d-survey.github.io
• Github: https://github.com/ziplab/Awesome-Feed-Forward-3D

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SkVM: Compiling Skills for Efficient Execution Everywhere

📝 Summary:
SkVM is a compilation and runtime system that enables portable and efficient execution of LLM skills across different models and platforms by treating skills as code and analyzing capability requireme...

🔹 Publication Date: Published on Apr 6

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.03088
• PDF: https://arxiv.org/pdf/2604.03088
• Project Page: https://skillvm.ai/index.html
• Github: https://github.com/SJTU-IPADS/SkVM

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Geometric Context Transformer for Streaming 3D Reconstruction

📝 Summary:
LingBot-Map is a feed-forward 3D foundation model that reconstructs scenes from video streams using a geometric context transformer architecture with specialized attention mechanisms for coordinate gr...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.14141
• PDF: https://arxiv.org/pdf/2604.14141
• Project Page: https://technology.robbyant.com/lingbot-map
• Github: https://github.com/robbyant/lingbot-map

🔹 Models citing this paper:
• https://huggingface.co/robbyant/lingbot-map

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Narrative-Driven Paper-to-Slide Generation via ArcDeck

📝 Summary:
ArcDeck is a multi-agent framework for paper-to-slide generation that models a paper's logical flow through discourse trees. It uses an iterative refinement process to ensure narrative coherence and improve presentations over direct summarization methods.

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11969
• PDF: https://arxiv.org/pdf/2604.11969
• Project Page: https://arcdeck.org/
• Github: https://github.com/RehgLab/ArcDeck

✨ Datasets citing this paper:
• https://huggingface.co/datasets/ArcDeck/ArcBench

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

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

#AIPresentations #NLP #GenerativeAI #ResearchTools #AcademicPublishing

✨HDR Video Generation via Latent Alignment with Logarithmic Encoding

📝 Summary:
This work enables high dynamic range HDR video generation by leveraging pretrained generative models. It uses logarithmic encoding to align HDR imagery with model latent spaces and camera-mimicking degradation training, achieving strong results without architectural redesign or complex retraining.

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11788
• PDF: https://arxiv.org/pdf/2604.11788
• Project Page: https://hdr-lumivid.github.io/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨LangFlow: Continuous Diffusion Rivals Discrete in Language Modeling

📝 Summary:
LangFlow demonstrates that continuous diffusion models can match discrete counterparts in language modeling by leveraging embedding-space flow matching with novel training techniques and noise schedul...

🔹 Publication Date: Published on Apr 15

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.11748
• PDF: https://arxiv.org/pdf/2604.11748
• Project Page: https://caradryanl.github.io/blog/2026/langflow/
• Github: https://github.com/nealchen2003/LangFlow

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Self-Distillation Zero: Self-Revision Turns Binary Rewards into Dense Supervision

📝 Summary:
Self-Distillation Zero trains a model to transform binary rewards into dense token-level self-supervision through dual-role training and on-policy self-distillation, achieving superior performance in ...

🔹 Publication Date: Published on Apr 13

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.12002
• PDF: https://arxiv.org/pdf/2604.12002

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research