Searcharxiv⌕ Search

arXiv subjects

Mahtab Movahhedrad

Publications and source records attributed to Mahtab Movahhedrad.

3 recordsLinked to original sources

When Does an AUROC Comparison Survive a Change in Population Mix?

Does an AUROC comparison survive a change in the proportions of tasks, populations, or image corruptions? A pooled-AUROC ranking of failure detectors or uncertainty scores describes one particular mix of groups. We study two fixed scores for a common binary target under mixtures of fixed group-conditional laws. Using the cross-group AUC decomposition, we prove that every rational symmetric quadratic comparison geometry is realizable by finite scalar-score distributions. Universal dominance is thus coNP-hard, and even exhaustive checks of every mixture on fewer than $K$ groups can miss a $K$-group reversal. A grid bound gives a dimension-uniform approximation certificate with sharp $1/m$ order. Common score marginals across groups make the comparison numerator linear, with an error envelope for imperfect alignment. These results yield exact finite-bank audits, explicit reversal mixtures and distances, and simultaneous population statements. For failure detection with a frozen ResNet-50 on ImageNet-C, 15 of 18 confidence-score comparisons win in every corruption family on a discovery split, and all 15 are finite-bank certified over all family mixtures on a held-out split; over all 15 corruption types, of 14 discovery winners the corrected grid certifies 10, leaves 3 unresolved, and finds one confirmation reversal of at least 0.05 AUROC. An audit of all 11,019 released Stable LM 2 12B outputs and 35 uncertainty methods finds 66 confirmation-set comparisons that win within all four tasks yet reverse under mixing. Discovery-fitted marginal rank alignment removes all 66; an alignment-error envelope certifies 83 comparisons. Label checks also change several selected reversals. The findings distinguish robustness of frozen scores from robustness after calibration; they do not establish human-error detection, robustness to unseen corruptions, or generalization beyond the evaluated banks.

cs.LG↗

Exact Sparse Assignment with Costly Target Embeddings: Fixed-Realization Certificates and an Empirical Audit

Embedding-based one-to-one association is common in vision and data integration: detections are associated with tracks and records are linked across databases. Such pipelines often solve a sparse optional assignment whose edge scores add a cheap term to an inner product of neural embeddings. When target embeddings are costly, one can acquire them selectively and stop once a certificate proves that the current assignment is optimal. We give such a certificate for unit-norm embeddings. It bounds the unknown part of every reassignment by a support function and maximizes the resulting robust gains with a second sparse assignment; a shared owner-centred bound dominates any fixed-centre triangle bound. We make precise what is certified when network features are not batch-invariant: the realized values of the acquired rows and every unit-ball completion of the rest, not a second forward pass in a different batch. Auditing against a separately batched cache tests a different property; on our first pilot it produced a spurious $2.49\times10^{-6}$ objective failure. We audit the corrected protocol on four public text entity-matching benchmarks with two frozen encoders; no image data is used. All 96 certificate simulations and 120 online selective timings were exact on their own realization (largest objective gap $6.82\times10^{-13}$). At group size 8 the shared certificate acquires on average 60% of candidate-active targets, against 74% (mean centre), 69% (MEB) and 60% (zero centre): it beats the mean and MEB centres but not the zero centre, and savings concentrate in one benchmark (7% acquired). The prespecified runtime claim is contradicted: selective execution is +0.510 s slower per run (dataset-cluster 95% interval [+0.248, +0.772]). Uniform-microbatch re-encodings preserved the optimum in only 5 of 8 encoder-dataset cells, and 45 of 120 online runs disagreed with an eager cache at near-ties.

cs.LG↗

Sharp First-Hit Identification from Finite Depth Samples

Which depths in a new camera view are determined by finitely many source-depth measurements? Pointwise uncertainty bounds do not answer this question: unobserved geometry between samples can change which surface is encountered first. We study a bounded Lipschitz class of continuous source inverse-depth fields with interval observations and known camera pose. Building on classical Lipschitz envelopes, we characterize the exact set of attainable first-hit depths and construct a globally admissible scene for every value in that set. The set can be disconnected. We also give a necessary and sufficient test for several proposed depths to be first hits of one shared scene. An epipolar reduction uses scalar cone inequalities without enumerating surfaces. In a rectified model with source spacing $h$, interval half-width $\varepsilon$, baseline $b$, and slope bound $L$, we prove a sharp worst-case inverse-depth diameter of $\max\{Lh+2\varepsilon,2\varepsilon/(1-bL)\}$ when $bL<1$, subject to sufficient depth and domain range for attainment. Exact samples still give the sharp bound $Lh$ at $bL=1$, while above this threshold ambiguity need not vanish as sampling becomes arbitrarily dense. Any positive measurement uncertainty can destroy uniform stability at the critical threshold. We provide constructive examples, an exact rational validation oracle, and a frozen ten-scene evaluation using reference, stereo, and learned-depth observations. The experiments separate solver correctness under the scene assumptions from failure of those assumptions: fixed learned-depth bands often yield narrow but incorrect sets. The results identify what a specified scene class permits; they do not infer a continuum regularity or physical-scene guarantee from sampled predictions.

cs.LG↗