✨Benchmarking and Mechanistic Analysis of Vision-Language Models for Cross-Depiction Assembly Instruction Alignment

📝 Summary:
Vision Language Models struggle with aligning assembly diagrams and video feeds due to a depiction gap, with findings indicating visual encoding as the primary target for improving cross-depiction rob...

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00913
• PDF: https://arxiv.org/pdf/2604.00913
• Project Page: https://ryenhails.github.io/IKEA-Bench/
• Github: https://ryenhails.github.io/IKEA-Bench/

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Understand and Accelerate Memory Processing Pipeline for Disaggregated LLM Inference

📝 Summary:
LLM inference faces significant memory processing overhead. This paper proposes using heterogeneous GPU-FPGA systems to accelerate these operations by offloading memory-bounded tasks to FPGAs. This achieves 1.04-2.2x speedup and 1.11-4.7x energy savings over GPU baselines, proving heterogeneous s...

🔹 Publication Date: Published on Mar 30

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.29002
• PDF: https://arxiv.org/pdf/2603.29002

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

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

#LLMInference #FPGA #HeterogeneousComputing #HardwareAcceleration #SystemArchitecture

✨UniMixer: A Unified Architecture for Scaling Laws in Recommendation Systems

📝 Summary:
UniMixer is a unified architecture for recommendation systems that improves scaling efficiency. It uses a generalized parameterized token mixing module to optimize mixing patterns and connect attention, TokenMixer, and factorization-machine methods. A lightweight version boosts performance further.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00590
• PDF: https://arxiv.org/pdf/2604.00590

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨ClawKeeper: Comprehensive Safety Protection for OpenClaw Agents Through Skills, Plugins, and Watchers

📝 Summary:
OpenClaw agents face critical security vulnerabilities due to extensive operational privileges. ClawKeeper provides comprehensive real-time protection using skill-based, plugin-based, and novel watcher-based mechanisms for state verification and intervention.

🔹 Publication Date: Published on Mar 25

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.24414
• PDF: https://arxiv.org/pdf/2603.24414
• Project Page: https://huggingface.co/datasets/xunyoyo/clawkeeper
• Github: https://github.com/SafeAI-Lab-X/ClawKeeper

✨ Datasets citing this paper:
• https://huggingface.co/datasets/xunyoyo/clawkeeper

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

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

#AISafety #AgentSecurity #AIagents #Cybersecurity #AIResearch

✨MemRerank: Preference Memory for Personalized Product Reranking

📝 Summary:
MemRerank improves personalized product reranking by distilling user purchase history into concise preference signals using reinforcement learning. This framework consistently outperforms raw history and other baselines, proving explicit preference memory is effective for e-commerce personalization.

🔹 Publication Date: Published on Mar 31

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.29247
• PDF: https://arxiv.org/pdf/2603.29247

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

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

#Personalization #ECommerce #ReinforcementLearning #RecommendationSystems #MachineLearning

✨Reasoning Shift: How Context Silently Shortens LLM Reasoning

📝 Summary:
LLMs significantly shorten their reasoning traces when problems are presented in various contexts compared to isolation. This compression reduces self-verification, potentially affecting performance on complex tasks. It highlights issues with LLM reasoning robustness and context management.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01161
• PDF: https://arxiv.org/pdf/2604.01161

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

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

#LLM #AIReasoning #ContextualAI #AIRobustness #MachineLearning

✨A Survey of On-Policy Distillation for Large Language Models

📝 Summary:
On-Policy Distillation OPD lets LLMs learn from self-generated outputs and teacher feedback, addressing off-policy exposure bias. This survey unifies OPD with an f-divergence framework, organizing methods by feedback, teacher access, and loss.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00626
• PDF: https://arxiv.org/pdf/2604.00626

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

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

#LLMs #OnPolicyDistillation #ModelDistillation #DeepLearning #MachineLearning

✨AI Generalisation Gap In Comorbid Sleep Disorder Staging

📝 Summary:
AI sleep staging models trained on healthy subjects perform poorly on stroke patients due to fundamental differences in sleep architecture. This necessitates disease-specific approaches. The paper introduces iSLEEPS, a new stroke dataset, to confirm this generalization gap and highlights the need...

🔹 Publication Date: Published on Mar 24

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.23582
• PDF: https://arxiv.org/pdf/2603.23582
• Project Page: https://himalayansaswatabose.github.io/iSLEEPS_Explainability.github.io/
• Github: https://github.com/HimalayanSaswataBose/iSLEEPS_GeneralisationGapAndExplainability

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

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

#AIGeneralization #SleepStaging #StrokeResearch #MedicalAI #MachineLearning

✨Brevity Constraints Reverse Performance Hierarchies in Language Models

📝 Summary:
Large language models can underperform smaller ones due to verbose responses that introduce errors. Constraining output length reveals their superior latent capabilities, reversing performance hierarchies. This demands scale-aware prompt engineering for optimal performance.

🔹 Publication Date: Published on Mar 11

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00025
• PDF: https://arxiv.org/pdf/2604.00025
• Github: https://github.com/logicsame/Brevity-Constraints-Reverse-Performance-Hierarchies-in-Language-Models

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

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

#LLM #PromptEngineering #AI #MachineLearning #NLP

✨Do Phone-Use Agents Respect Your Privacy?

📝 Summary:
This paper introduces MyPhoneBench, a framework to evaluate phone agents' privacy behavior. It found agents often over-share optional data, indicating current success metrics overestimate their deployment readiness due to privacy failures.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00986
• PDF: https://arxiv.org/pdf/2604.00986
• Github: https://github.com/FreedomIntelligence/MyPhoneBench

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

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

#PhoneAgents #DataPrivacy #AI #PrivacyResearch #Cybersecurity

✨S0 Tuning: Zero-Overhead Adaptation of Hybrid Recurrent-Attention Models

📝 Summary:
S0 tuning optimizes recurrent state matrices in hybrid models, outperforming LoRA with zero inference overhead. It significantly improves performance on benchmarks like HumanEval and enables efficient task switching.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01168
• PDF: https://arxiv.org/pdf/2604.01168
• Project Page: https://www.jackyoung.io/research/s0-tuning
• Github: https://github.com/JackYoung27/s0-tuning

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

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

#S0Tuning #DeepLearning #LLMs #ModelOptimization #MachineLearning

✨When Users Change Their Mind: Evaluating Interruptible Agents in Long-Horizon Web Navigation

📝 Summary:
LLM agents struggle with user interruptions during long web navigation tasks. This paper introduces InterruptBench, the first systematic study and benchmark to evaluate interruptible agents in these scenarios. Results show that current LLMs find handling mid-task interruptions effectively and eff...

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00892
• PDF: https://arxiv.org/pdf/2604.00892
• Project Page: https://github.com/HenryPengZou/InterruptBench
• Github: https://github.com/HenryPengZou/InterruptBench

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

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

#LLMAgents #UserInteractions #WebNavigation #AIResearch #Benchmarking

✨Consistency Amplifies: How Behavioral Variance Shapes Agent Accuracy

📝 Summary:
Behavioral consistency in LLM agents correlates with higher accuracy across models. However, consistency can amplify both correct and incorrect interpretations, meaning consistent wrong interpretations are a major failure mode. Thus, accurate interpretation is more crucial than execution consiste...

🔹 Publication Date: Published on Mar 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.25764
• PDF: https://arxiv.org/pdf/2603.25764

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

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

#LLMAgents #ModelAccuracy #BehavioralAI #AIResearch #AIInterpretation

✨AgentWatcher: A Rule-based Prompt Injection Monitor

📝 Summary:
AgentWatcher defends LLMs against prompt injection, which struggles with long contexts and opaque detection. It achieves scalability by using causal attribution to pinpoint influential context segments. Detection is explainable through a monitor LLM that applies explicit rules.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01194
• PDF: https://arxiv.org/pdf/2604.01194
• Github: https://github.com/wang-yanting/AgentWatcher

🔹 Models citing this paper:
• https://huggingface.co/SecureLLMSys/AgentWatcher-Qwen3-4B-Instruct-2507

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨PixelPrune: Pixel-Level Adaptive Visual Token Reduction via Predictive Coding

📝 Summary:
PixelPrune reduces VLM computational costs by removing redundant image patches before Vision Transformer encoding. It uses predictive-coding compression in pixel space, speeding up inference and training up to 4.2x and 1.9x respectively while maintaining accuracy.

🔹 Publication Date: Published on Apr 1

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.00886
• PDF: https://arxiv.org/pdf/2604.00886
• Github: https://github.com/OPPO-Mente-Lab/PixelPrune

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨GPA: Learning GUI Process Automation from Demonstrations

📝 Summary:
GUI Process Automation (GPA) offers robust, deterministic, and privacy-preserving vision-based robotic process automation with faster execution than current vision-language model approaches. AI-genera...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.01676
• PDF: https://arxiv.org/pdf/2604.01676
• Project Page: https://www.salesforceairesearch.com/gpa

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨The Latent Space: Foundation, Evolution, Mechanism, Ability, and Outlook

📝 Summary:
Latent space is emerging as a fundamental computational substrate for language-based models, offering advantages over explicit token-level approaches through continuous representation that mitigates l...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02029
• PDF: https://arxiv.org/pdf/2604.02029
• Github: https://github.com/YU-deep/Awesome-Latent-Space

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨SKILL0: In-Context Agentic Reinforcement Learning for Skill Internalization

📝 Summary:
SKILL0 enables LLM agents to internalize skills during training, allowing zero-shot autonomous behavior through a dynamic curriculum that reduces contextual overhead while improving task performance. ...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02268
• PDF: https://arxiv.org/pdf/2604.02268
• Github: https://github.com/ZJU-REAL/SkillZero

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨FlowSlider: Training-Free Continuous Image Editing via Fidelity-Steering Decomposition

📝 Summary:
FlowSlider enables continuous image editing with slider-style control by decomposing updates into fidelity and steering components within Rectified Flow, providing stable strength control without addi...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02088
• PDF: https://arxiv.org/pdf/2604.02088
• Project Page: https://huggingface.co/spaces/dominoer/FlowSlider

✨ Spaces citing this paper:
• https://huggingface.co/spaces/dominoer/FlowSlider

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨DataFlex: A Unified Framework for Data-Centric Dynamic Training of Large Language Models

📝 Summary:
DataFlex is a unified framework for dynamic data-centric training of large language models that supports sample selection, domain mixture adjustment, and sample reweighting while maintaining compatibi...

🔹 Publication Date: Published on Mar 27

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2603.26164
• PDF: https://arxiv.org/pdf/2603.26164
• Github: https://github.com/OpenDCAI/DataFlex

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research

✨Generative World Renderer

📝 Summary:
A large-scale dynamic dataset derived from AAA games is introduced to improve generative inverse and forward rendering, featuring high-resolution synchronized RGB and G-buffer data alongside a novel V...

🔹 Publication Date: Published on Apr 2

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.02329
• PDF: https://arxiv.org/pdf/2604.02329
• Project Page: https://alaya-studio.github.io/renderer
• Github: https://github.com/ShandaAI/AlayaRenderer

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

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

#AI #DataScience #MachineLearning #HuggingFace #Research