Search papers, labs, and topics across Lattice.
This study investigates the significance of representation trajectories in vision models, revealing that the computation paths leading to final representations contain valuable information that enhances out-of-distribution (OOD) detection and image classification. By analyzing the transformations across model depth while maintaining sample identity, the authors identify distinct patterns of continuity and architecture-specific depth profiles that improve performance on various datasets. The findings demonstrate that a novel transition-surprise score can significantly reduce false positive rates in OOD detection, particularly in challenging scenarios involving visually disruptive shifts.
Representation trajectories can provide a critical reliability signal that enhances OOD detection and classification performance, outperforming traditional final-state approaches.
Vision models do not form a representation at once; each block revises it. We ask whether the resulting computation path contains evidence that the final representation discards, and whether that evidence improves OOD detection and image classification on clean and shifted data. Unlike approaches that treat intermediate layers as separate snapshots, we retain sample identity across depth and study the transformations connecting successive states. We separate class-coherent transport from input-specific innovation, and coordinate movement from relational reorganization. Across supervised, self-supervised, vision--language, hierarchical, and convolutional encoders, these paths show strong sample-specific continuity and architecture-specific depth profiles that recur across datasets. They are also practically useful. An ID-only transition-surprise score complements strong final-state detectors, reducing FPR95 in 131/152 non-saturated comparisons on a balanced OpenOOD grid; gains are largest for visually disruptive and semantically far shifts, and remain positive on near-OOD for most detectors. Frozen update probes improve 71/72 clean model--dataset cases, while shifted-data gains vary with architecture and corruption type. Computation paths therefore provide a broadly useful reliability signal whose value is determined jointly by model organization and the shift encountered.
