Your daily OSS contribution digest

This paper presents a new consensus protocol discovered with AI assistance that significantly improves latency in geo-distributed systems, and introduces an AI-driven framework for autonomously designing and verifying distributed protocols.

• Semantic Observability for Autonomous Systems Diagnostics

  This paper presents an autonomous system that integrates diverse logs into a unified graph model and uses multi-agent algorithms plus large language models to automatically diagnose and explain failures in complex distributed systems.

  Ai Ml · relevance 9/10
• Train An Imaginative Agent

  This paper presents a training framework that enables an AI agent to imagine and simulate future outcomes of its actions before executing them, improving its planning and decision-making in complex environments like Sokoban and Pacman.

  Ai Ml · relevance 9/10
• Evolving Agent Interfaces

  This paper presents Agent S4, an AI system that autonomously creates and improves its own interfaces for computer tasks, significantly boosting performance beyond fixed-interface models and even outperforming larger, more costly AI systems.

  Ai Ml · relevance 9/10

• Pre-emptive Carbon Aware Scheduling for Edge Workloads

  This paper develops efficient algorithms to schedule edge computing jobs in a way that reduces carbon emissions by shifting workloads to cleaner energy periods, balancing timing constraints and cloud outsourcing costs.

  Algorithms Theory · relevance 8/10
• Fully Dynamic Triangle and 4-Vertex Subgraph Counting: From Theory to Practice and Back Again

  This paper improves and evaluates dynamic algorithms for counting triangles and four-vertex subgraphs in large evolving graphs, achieving significant speedups and providing practical guidance for their use.

  Algorithms Theory · relevance 8/10
• Temporal Exploration and Rendezvous via Weisfeiler-Leman Colorings

  This paper presents efficient, deterministic algorithms to explore and rendezvous in temporal graphs much faster than previous methods, using a graph coloring technique to reduce complexity.

  Algorithms Theory · relevance 7/10

• Building SQL Equivalence Proofs from Query Rewrite Rules

  This paper presents a system that uses SQL query rewrite rules to construct formal proofs that two SQL queries are equivalent, improving reliability and interpretability of query rewriting and text-to-SQL semantic evaluation.

  Database · relevance 8/10
• WaRP: Workload-Aware Request Planning for Vector Retrieval on Object Storage

  This paper introduces WaRP, a method that optimizes how vector search systems group data requests to cloud object storage, significantly improving query speed and reducing costs without losing accuracy.

  Database · relevance 8/10

• HAWK: Heterogeneity-Aware KV Cache Transfer for Efficient GPU-NPU Phase-Disaggregated LLM Inference

  This paper presents HAWK, a method to efficiently transfer key-value cache data between GPUs and NPUs during large language model inference, enabling better hardware utilization and improved power and cost efficiency.

  Distributed Systems · relevance 8/10
• Weight Placement Is Not Enough: Synchronization in Multi-Node CPU Decoding of Ternary LLMs

  This paper identifies synchronization delays as a key bottleneck in multi-node CPU decoding of ternary large language models and demonstrates a new thread partitioning strategy that significantly speeds up decoding while reducing energy use.

  Distributed Systems · relevance 8/10

• Scaling H2 Factorization to Billion-Unknown Three-Dimensional Problems

  This paper introduces a scalable distributed solver for very large 3D integral equation problems, achieving efficient factorization of matrices with billions of unknowns by combining novel algorithmic and parallelization techniques.

  Math · relevance 8/10
• Swarmlet: a GPU-resident reformulated vortex particle method for coupled multirotor aerodynamics and flight dynamics

  This paper introduces Swarmlet, a GPU-based simulation tool that accurately models the aerodynamics and flight dynamics of multiple multirotor drones together, enabling efficient and precise swarm-scale analysis.

  Math · relevance 8/10
• When Windowed Anderson Acceleration Converges Faster

  This paper precisely characterizes when a technique called windowed Anderson acceleration speeds up convergence for solving linear fixed-point problems, providing explicit criteria and convergence rate bounds.

  Math · relevance 7/10

• VX-One: Typed, Calibrated Decision Heads for Real-Time Perimeter Triage - An empirical study of unknown-attack honesty, cascade gray zones, and calibration under drift

  This paper introduces VX-One, a fast multi-head neural network producing calibrated, typed decision outputs for real-time network perimeter triage, and empirically evaluates its detection performance, calibration, and robustness under temporal drift against established methods.

  Networking · relevance 8/10
• Enhancing IoT Network Reliability through Antenna Diversity in Real-World Scenarios

  This paper shows that using multiple antennas with different polarizations on IoT devices can greatly improve wireless signal strength and network reliability in practical environments.

  Networking · relevance 8/10
• SAKURA: Serving Attention Key-Value State Under Rate Adaptation for Private Image Editing on Phones

  This paper presents SAKURA, a system that efficiently delivers reusable attention key-value states from the cloud to mobile devices for private image editing, adapting to network conditions to reduce latency and storage while preserving image quality.

  Networking · relevance 8/10

• GhostFaceMPC: Privacy-Preserving Face Recognition with Multi-Party Computation

  This paper introduces GhostFaceMPC, a system that enables secure face recognition by processing facial data privately using multi-party computation, preventing exposure of sensitive biometric information throughout the process.

  Security · relevance 9/10
• Temporal Dynamics and Security Assessment of SRT Protocol: A 96-Day Longitudinal Analysis of Production Streaming Infrastructure

  This paper analyzes the behavior and security of the SRT streaming protocol over 96 days in a real-world setting, revealing how system resources relate to transport metrics and demonstrating effective anomaly detection methods for security events.

  Security · relevance 8/10
• Prefilling the Reasoning Channel: Output-Prefix Attacks on Reasoning LLMs

  This paper demonstrates a new type of prompt injection attack on reasoning-capable large language models by manipulating their intermediate reasoning steps combined with output prefixes, achieving very high success rates in bypassing model safeguards.

  Security · relevance 8/10