Triscore combines three signals to detect unknown inputs robustly
TriScore: Post-Hoc Out-of-Distribution Detection with Energy, Boundary Probes, and Transform Consistency
Computer Vision and Pattern RecognitionMachine Learning
Summary
Sometimes, AI models can be tricked by inputs they haven't seen before, which can cause mistakes. The authors found that relying on just one way to detect these surprises can fail if the model is slightly changed. They built a method called TriScore that looks at three different clues from a model to better spot when something is unfamiliar. TriScore works well without needing extra training or examples of strange inputs and is more reliable under tricky conditions.
What this means in practice
- •For machine learning engineers: Improve system reliability by integrating TriScore to detect unusual inputs that might cause AI model errors without retraining or extra resources.
- •For security operations teams: Enhance defenses by catching unexpected or adversarial inputs that evade single-signal detectors using TriScore’s combined criteria.
Authors
Ege Gursoy, Nicolas Mansard
Abstract
Post-hoc out-of-distribution detection usually relies on a single signal extracted from a frozen classifier, such as confidence, energy, features, or gradients. This dependence becomes fragile under classifier-level adversarial shift, where the shift can preserve the detector's chosen cue while changing the model's decision behavior. We investigate whether this failure mode can be mitigated by combining complementary detection criteria instead of selecting a single score. We introduce TriScore, a post-hoc OOD detector that combines three ID-calibrated signals from the same frozen backbone: an energy prior, a boundary-fragility score, and a transform-consistency score. These signals are standardized on ID validation data and fused by an input-dependent ID-residual gate, requiring no OOD labels, auxiliary model, training, or stored feature bank. Across four ImageNet-scale backbones, TriScore achieves the best mean AUROC among 15 post-hoc baselines under a non-adaptive adversarial-shift protocol, while remaining competitive on standard OOD benchmarks. Ablations show that the gains come from combining complementary cues rather than optimizing a single detector signal. These results support multi-criterion post-hoc detection as a practical route toward more robust OOD detection under classifier-level shift. The source code is publicly available at github.com/gursoyege/TriScore.