Benchmark RadarST-Bench
ST-Bench is a verification benchmark for evaluating certified robustness of spatio-temporal neural networks on autonomous driving (Udacity) and activity recognition (UCF-101) tasks, using spatio-temporal perturbation constraints.
- Released
- 2026-06-08
- Readiness
- Paper only
- Primary field
- Transport & Logistics
Why it matters
Existing robustness verification methods rely on overly conservative assumptions or are computationally prohibitive. ST-Bench provides a realistic, constrained perturbation model for video inputs, enabling tighter approximations and more meaningful robustness comparisons for safety-critical applications.
Motivation
With AI increasingly deployed in safety-critical systems, providing formal robustness guarantees for the underlying models is essential.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.