AI/ML arXiv cs.AI

Field-Aware RankMixer with Dual-Stream Bilinear Fusion for the Tencent UNI-REC Challenge

The authors present FA-RankMixer, a model for multi-domain user behavior sequence modeling for ad pCVR prediction, ranking ninth in the Tencent UNI-REC Challenge.

AI/ML arXiv cs.AI

Scalable LLM Agent Tool Access in the Cloud

A cloud-scale gateway system is proposed for the Model Context Protocol (MCP) to solve tool selection latency and scalability issues for LLM agents.

AI/ML arXiv cs.AI

Process Reward Informed Tree Rollout for Effective Multi-Turn RL

PATR is a rollout framework for multi-turn agent RL that uses process-guided tree rollouts to improve exploration efficiency on benchmarks like SWE-Bench.

Open Source arXiv cs.AI

AEGIS: Assay-Aware Protocol Validation and Runtime Monitoring for Open-Source Liquid Handling Robots

AEGIS is an open-source system that provides assay-aware protocol validation and visual runtime monitoring for liquid handling robots like the Opentrons OT-2.

AI/ML arXiv cs.AI

Think at 5 Hz, Act at 20 Hz: Asynchronous Fast-Slow Vision-Language-Action Inference for Closed-Loop Driving

A fast-slow architecture for closed-loop driving is introduced, decoupling low-frequency scene reasoning from high-frequency action execution to reduce latency.

Software Engineering arXiv cs.AI

A cubical formalisation of topos causal models: intervention, sheaf gluing, and the intuitionistic do-calculus

A machine-checked formalisation of topos causal models is implemented in Cubical Agda, providing a verified account of intervention and sheaf gluing.

AI/ML arXiv cs.AI

IMBench: A Benchmark for Intuitive Robotic Manipulation

IMBench is a new benchmark designed to evaluate the integrated physical reasoning and motor control capabilities of intuitive robotic manipulation.

Other Hacker News

Sealed Tomb Filled with Paintings and Inscriptions Discovered in Egypt

Discovery of a sealed tomb containing paintings and inscriptions in Egypt.

Software Engineering Hacker News

How proprietary formats have become Microsoft’s main tool for lock-in

An analysis of how Microsoft uses proprietary file formats as a mechanism for vendor lock-in.

AI/ML arXiv cs.AI

Cache-Aware Prompt Compression:A Two-Tier Cost Model for LLM API Caching

Introduces Cache-Aware Prompt Compression (CAPC) to optimize LLM API costs by balancing prompt caching and compression strategies.

AI/ML arXiv cs.AI

SLAPBench: Benchmarking Multimodal Large Language Models for Four-Finger SLAP Fingerprint Verification

Presents SLAPBench, a benchmark for evaluating Multimodal LLMs on four-finger SLAP fingerprint verification.

AI/ML arXiv cs.AI

Recursive Harness Self-Improvement

Introduces Recursive Harness Self-Improvement (RHI) to optimize agent loops and improve execution-trace quality for model training.

AI/ML arXiv cs.AI

Kolmogorov--Arnold Networks for Small Language Models

Evaluates Kolmogorov-Arnold Networks (KANs) as potential replacements for MLPs in small language models, finding limited advantage over strong baselines.

AI/ML arXiv cs.AI

CoWeaver: A Bi-directional, Learnable and Explainable Matching Engine for Mixed Human-Agent Science Collaboration

Proposes CoWeaver, a learnable matching engine designed to facilitate collaboration between human scientists and AI agents.

AI/ML arXiv cs.AI

From Feasibility to Desirability: Plan, Learn, Adapt (PLA) Framework for Personalized On-Device Itinerary Generation

Presents the PLA framework for personalized on-device itinerary generation, combining lightweight planners and a Bradley-Terry reward model.

AI/ML arXiv cs.AI

Evolutionary Algorithm-Guided LLMs for Physics-Informed Neural Network Design

Develops a closed-loop evolutionary algorithm that guides LLMs to design optimal configurations for Physics-Informed Neural Networks (PINNs).

AI/ML arXiv cs.AI

Hard Rules, Soft Preferences: Bridging Reasoning, Learning, and Optimization for Personalized Packing Checklist Generation

Introduces a reasoning-guided learning framework for personalized packing checklists using a symbolic engine and CP-SAT optimizer.

Other Hacker News

Self-Powered Trailers Promise Leaner Freight Runs

Article discussing self-powered trailers designed to reduce fuel consumption and emissions in freight transport.

Hardware/Chips Hacker News

Xiaomi-Robotics-1

Discussion regarding Xiaomi's robotics initiatives.

AI/ML arXiv cs.AI

Partial Information Decomposition as a Multi-Contrast 3D MRI Selection Strategy for Resource-Constrained Deep Neural Network Training in Brain Tumor Segmentation

A study on using Partial Information Decomposition to select the most informative MRI input pairs for efficient brain tumor segmentation in deep learning.