Your daily OSS contribution digest

This paper presents 3DCodeVerse, a comprehensive system and dataset that enables AI agents to generate, verify, and refine 3D content through executable code, significantly improving automated 3D modeling capabilities.

• WeAct: Dynamic Weight Switching as Agent Actions

  This paper introduces a method that lets AI agents dynamically switch between specialized internal modules during tasks, improving performance and enabling modular updates without retraining the entire model.

  Ai Ml · relevance 8/10
• OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding

  This paper introduces OmniCompass, a comprehensive benchmark to evaluate multimodal large language models on video understanding tasks involving vision, language, and audio, revealing current models' limitations in precise temporal and spatial grounding.

  Ai Ml · relevance 8/10

• The Value of Experiments in Assortment Optimization

  This paper analyzes how retailers can use limited product assortment experiments to better learn customer preferences and improve revenue, showing that even a single well-chosen experiment can significantly reduce potential losses compared to no experimentation.

  Algorithms Theory · relevance 7/10
• Half-Optimal Induced Matchings in Random Graphs: Online and Low-Degree Algorithms

  This paper shows that for finding large induced matchings in random graphs, simple online and low-degree algorithms can only achieve about half the optimal size, and this limitation is provably sharp even with powerful algorithmic models.

  Algorithms Theory · relevance 7/10
• One Polynomial Field for Every Euler Step and the Flow: Polynomial Dynamics for P and PSPACE

  This paper shows how to represent any deterministic computation as a single polynomial vector field that simulates the computation both in discrete Euler steps and continuously, unifying discrete and continuous models of computation and characterizing complexity classes like P and PSPACE through polynomial dynamics.

  Algorithms Theory · relevance 7/10

• Can We Rely on Language Models to Orchestrate Services across the Edge-Cloud Continuum?

  This paper reviews how language models are currently used to help manage and coordinate services between edge and cloud computing, highlighting where they are reliable and where safeguards and human oversight remain necessary.

  Distributed Systems · relevance 7/10
• AI in the EGI e-infrastructure -- Past experiences and future strategies

  This paper reviews the European Grid Infrastructure's experience with AI in scientific projects and outlines their strategic plan to build AI-ready infrastructure, develop human expertise, and ensure trustworthy AI deployment for research by 2030.

  Distributed Systems · relevance 6/10
• SPECTRUM: A Strategic Framework and Technical Blueprint for European Exascale Research Data and Compute Infrastructure

  This paper outlines a strategic and technical plan to build a European data and computing infrastructure capable of handling exabyte-scale scientific data from major physics and astronomy projects over the next decade.

  Distributed Systems · relevance 6/10

• Graph-Convolutional Reinforcement Learning for Finite-Time Stabilization of a Chaotic Chua Circuit via Unified Flexible Terminal Sliding-Mode Control

  This paper presents a method that uses graph convolutional reinforcement learning to automatically tune a robust controller, enabling fast and reliable stabilization of chaotic electrical circuits despite uncertainties and input limits.

  Math · relevance 7/10
• Integer-Quantized Hierarchical Matrices with Adaptive Bit Allocation

  This paper introduces a method to efficiently compress hierarchical matrices by adaptively assigning different bitwidths to their low-rank components, significantly reducing storage while controlling reconstruction error.

  Math · relevance 7/10
• Perfect sampling of Ising spin glasses with bounding chains

  This paper develops and analyzes algorithms for perfectly sampling configurations of Ising spin glasses, showing efficient performance in certain graph regimes and identifying limitations that suggest new directions for improved sampling methods.

  Math · relevance 7/10

• Prompt Injection Through Forensic Evidence: A Taxonomy and Proof of Concept of Anti-Forensic Attacks on LLM-Assisted Digital Investigations

  This paper shows how attackers can manipulate large language models used in digital forensic investigations by embedding malicious prompts in evidence, causing false conclusions or harmful actions, and proposes ways to defend against such attacks.

  Security · relevance 8/10
• Asynchronous Verifiable and Complete Secret Sharing: From Design Patterns to System-Level Trade-offs

  This paper systematically compares and evaluates different asynchronous verifiable secret sharing protocols, providing practical guidance on their performance trade-offs and implementation choices.

  Security · relevance 8/10
• Weaponizing the Exit: Input Privacy Leakage via Exit Routing Manipulation in Multi-Exit Networks

  This paper reveals a new privacy risk where cloud providers can infer sensitive input properties by manipulating multi-exit neural networks and observing inference latency, and proposes a training method to mitigate this threat.

  Security · relevance 8/10