Research

MobiFlow: Real-World Mobile Agent Benchmarking through Trajectory Fusion

MobiFlow introduces trajectory fusion to benchmark mobile agents more rigorously, advancing evaluation methodology for AI systems interacting with real-world mobile UIs.

Tuesday, April 14, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.AIBY sys://pipeline

MobiFlow presents a benchmarking framework for evaluating mobile agents — AI systems that interact with mobile UIs in real-world scenarios. The work uses trajectory fusion, a method for combining execution paths to improve agent evaluation. This contributes to understanding and measuring mobile agent capabilities.

Read original at arXiv CS.AI

How Stripe Detects Fraudulent Transactions Within 100 ms

Stripe's Shield NeXt fraud detector cuts training time 85% with a ResNeXt-inspired multi-branch architecture, processing 1,000+ signals per transaction in under 100ms while preserving a critical 1.5% recall edge through network effects and dark web intelligence.