A prediction that is both confident and wrong is a critical reliability failure because it can bypass abstention and human review precisely when the model is mistaken. Empirical risk minimization (ERM) controls average loss but not this failure directly, while calibration, uncertainty estimation, conformal risk control, and selective prediction methods target related reliability properties rather than bounding the joint failure event during training. We propose ReliableNet, which constrains the Joint Confident-Wrong (JCW) probability, the probability that a prediction is simultaneously confident and incorrect, below a user-specified risk budget $α\in(0,1)$. We formulate this as a chance-constrained ERM problem, use a conservative smooth inner approximation whose population feasibility implies the original JCW constraint. Across four tabular and two image datasets, ReliableNet is the only method certified within the JCW budget for every dataset and seed in distribution, when compared against baselines spanning ERM, post-hoc calibration, conformal risk control, and selective prediction. Under demographic, ambiguity, spurious-correlation, novel-class, and covariate shifts, it achieves the lowest empirical JCW among the compared methods while remaining very competitive in accuracy, coverage, calibration, and selective prediction. Risk-coverage results further indicate that ReliableNet achieves better selective ranking than the benchmark methods on most datasets. Overall, ReliableNet provides a principled approach to trustworthy classification.
Per-field accept/review with selective risk at most alpha -- accept a field only if the error rate among accepted fields is controlled -- is the trust contract document-extraction systems need, and the natural procedure silently violates it on real documents. On 13,859 genuine claude-sonnet-5 fields from 800 CORD receipts (49.0% correct) we diagnose three failure modes: document clustering (design effect 1.84-2.45), score-refit leakage (coverage 0.416 at risk 0.127, violating alpha=0.10 in 95% of splits), and a tie-mass pathology (a degenerate score collapses the threshold grid, 0.030 to 0.001). We organize the fixes as a validity ladder, guarantee form stated per tier. A fit/val split protocol restores expected-selective-risk control for a learned fusion: coverage 0.318 at risk 0.096 at nominal alpha=0.10, no tolerance band (production variant 0.326) -- an on-average point whose realized risk exceeds alpha in 47.5% of resplits, not a certificate. Mondrian Learn-then-Test with exact binomial tails yields per-group PAC certificates: field-iid 0.171 at risk 0.068, cluster-corrected 0.140, doc-iid 0.060 -- the only tier matching documents, honestly near-vacuous today. Support-bin, the pre-specified provenance taxonomy, wins every rigor tier on the sonnet CORD capture (p<1e-4, Bonferroni-corrected) -- a win that does not replicate on the same documents under haiku or qwen -- while on higher-accuracy corpora pooled thresholds win: conditioning helps exactly where pooled cannot certify, subsumed by a learned score elsewhere. A frozen-configuration confirmation on selection-untouched claude-haiku-4-5 held at both risk levels, and a blind three-annotator human-gold audit verifies the practical tier's accepted-set risk at 1.3% against its 10% budget (Fleiss' kappa=0.83; labels err one-sidedly pessimistic). Released Apache-2.0 with seed-pinned, regression-gated procedures.
Jakub Paplhám, Willem Waegeman, Eyke Hüllermeier +1cs.LG cs.AI
Current evaluation of epistemic uncertainty relies on tasks such as out-ofdistribution detection and active learning. However, the Bayes-optimal decision strategies for these tasks do not coincide with the scores commonly used to quantify epistemic uncertainty. Building on the epistemic reject-option framework, we evaluate epistemic uncertainty using its ability to identify regret, the reducible error. Formulating selective prediction as a constrained optimization over coverage, expected risk, and regret, we prove the optimal selector is a thresholded convex combination of the ground-truth aleatoric and epistemic uncertainties. This theoretical unification exposes a weakness in recent uncertainty disentanglement literature: we demonstrate that standard correlation metrics between learned components do not necessarily predict their actual operational utility. We instead propose to evaluate the achievable risk, regret, coverage surface of the decomposition as a diagnostic for joint disentanglement and utility. Benchmarking standard methods on datasets with dense human annotations reveals that decision-theoretic rankings can disagree substantially with proxy-task rankings, including pairwise rank inversions between methods that are top-ranked on one criterion and bottom-ranked on other.
In selective deployment, practitioners act only on a model-chosen subset of individuals based on predicted conditional average treatment effects, but marginal conformal guarantees need not control reliability on that selected subset. We study reliable selection for black-box CATE predictors: selecting candidates whose CATE errors are below a tolerance while controlling the false discovery rate (FDR). Since CATE errors are unobservable, we construct doubly robust proxy errors from pseudo-outcomes; however, naive proxies can lose power under heteroskedasticity because variance overwhelms the reliability signal. We propose Denoised Conformal Alignment, which subtracts an estimated conditional variance component and combines conformal calibration with Benjamini--Hochberg selection. Our analysis shows that validity is governed by stability of proxy/oracle threshold labels, rather than pointwise perfection of the variance estimator. Experiments show substantially improved power while maintaining FDR control across challenging settings.
High test accuracy and good aggregate calibration do not show whether an individual prediction is structurally supported by its evidence. In tabular decision systems, failures often occur when a feature family becomes unavailable, delayed, noisy, stale, or low-trust while the model remains highly confident. Existing calibration, uncertainty, selective-prediction, explanation, and perturbation methods provide scalar scores or attribution maps, but not a recomputable audit object answering: under a declared evidence-failure protocol, what trajectory makes this prediction lose support? We introduce Counterfactual Fragility Certificates (CFC), a model-agnostic protocol-level audit certificate-not a formal robustness certificate-that maps each prediction into an ordered evidence-failure trajectory summarized by greedy flip budget, normalized margin-collapse area, degradation thresholds, and fragility dominance score. Across seven tabular benchmarks and strong linear, tree-based, boosting, and neural baselines, CFC-FDS identifies independently brittle high-confidence cases with 0.915 AUROC, improving over the strongest non-certificate score by +0.405. The advantage persists across perturbation, permutation-importance, group-SHAP, baseline-choice, seed-variance, budgeted-review, and naturalistic field-unavailability checks. Under a 20% review budget, CFC-FDS captures 88.9% of brittle high-confidence cases, compared with 31.8-37.4% for confidence and energy scores. We also evaluate fragility-aware regularization and brittleness-aware temperature correction as secondary uses. CFC provides a concrete reliability framework for exposing high-confidence brittleness missed by ordinary score-centric evaluation.
Post-hoc calibration for time-series classification usually remaps output scores, but deployment decisions such as trust, abstention, and review depend on whether a confident prediction is supported by the current temporal signal. We address three time-series reliability gaps: identical confidence values can hide different temporal support, average calibration can miss false high-confidence errors, and output-space recalibration offers limited input-linked auditability. We introduce a validation-gated fixed-label reliability policy that keeps the backbone prediction unchanged while estimating whether it should be trusted. The method combines output-side cues with whole-sample spectral descriptors, including band energy, entropy, peak dominance, period support, and phase stability, to form a scalar reliability estimate and diagnostic band-level evidence. A validation gate enables spectral conditioning only when correctness ranking improves without breaching FalseConf@0.9 or AURC tolerances; otherwise it reverts to the safer output-space baseline. Across eight heterogeneous UCR/UEA datasets, eight time-series backbone families, and standard recalibrators, the unconstrained method improves fixed-label selective-reliability metrics on the matched evaluation subset, raising Corr-AURC from 0.693 to 0.779. The validation-gated policy further improves Corr-AURC to 0.786 and reduces FalseConf@0.9 to 0.094. These results suggest that reliability estimation for time-series classifiers benefits from bundling output confidence with spectral evidence, while validation gating prevents unsupported spectral conditioning.
Deep learning methods have achieved state-of-the-art in time series forecasting, yet their accuracy varies considerably across samples, as some instances remain inherently difficult to predict. Reject option mechanisms, which allow models to abstain from high-risk predictions, are well established in classification and regression but underexplored in forecasting. Existing abstention strategies typically rely on proxies, such as the width of the prediction interval or learned confidence scores derived from forecasts. However, these approaches are inherently tied to the training domain, limiting their ability to generalize. We propose a selective forecasting framework that addresses this limitation by modeling the empirical percentile of forecasting errors, that is, a scale-invariant statistic, based on structural characteristics extracted from recent lags via metalearning. By decoupling the rejection decision from the forecast itself and grounding it in domain-agnostic features, the framework enables effective abstention transfer across heterogeneous time series. Experiments in both in-domain and transfer learning settings show that rejecting samples predicted as challenging consistently improves forecasting accuracy across coverage levels.
Uncertainty quantification (UQ) is essential for reliable decision-making in safety-critical applications in probabilistic machine learning. For regression problems, dominant scalar UQ approaches - notably, those based on proper scoring rules - measure uncertainty via pointwise predictive risk. This can lead to counterintuitive results when the target statistic is not the conditional expectation. We propose an alternative framework, in which uncertainty is characterised by the volume of the most probable subset of a distribution's support. QUEST (Quantifying Uncertainty via highest dEnSiTy regions) is a novel approach to UQ based on the concentration of Lebesgue measure at a distribution's peak(s), evaluated at one or more values of a robustness parameter $α$. We establish connections between our measures and classical statistics from information theory and economics. We show that, unlike popular alternatives based on proper scoring rules, QUEST measures of epistemic and aleatoric uncertainty satisfy a set of axioms adapted from the UQ literature, including monotonicity under distributional spread and invariance to location shifts. Selective prediction benchmarks confirm that QUEST performs favourably against standard measures such as variance and differential entropy.
Selective prediction with distribution-free risk control promises that, with confidence 1-delta over the calibration draw, the error rate of accepted inputs stays below a user budget alpha. We audit this promise on signal-domain detectors -- machine anomalous-sound detection (ASD) and AI-generated-image forensics -- for four calibration rules: uncertified empirical thresholding (NAIVE) and certified Hoeffding, Clopper-Pearson (CP), and betting (WSR) upper confidence bounds. We report three findings. (i) NAIVE thresholding, common in practice, exceeds its declared budget in 49-73% of synthetic trials (n=200 calibration points) and in up to 68% of real-data splits: a false sense of safety rather than a broken theorem, since the rule never had a certificate. (ii) Tightness matters: CP and WSR certify substantial coverage where Hoeffding certifies none, with zero observed budget overruns under exchangeable splits. (iii) Under grouped deployment (unseen machine types or generators), certified rules overrun in 9-30% of trials -- far above delta -- showing the failure lies in the broken exchangeability premise, not in the bounds; a conservative per-group threshold restores validity at a severe coverage cost.
Selective predictors answer on confident inputs and abstain elsewhere; deploying one safely needs a single finite-sample certificate that simultaneously upper-bounds the selected risk, lower-bounds the acceptance probability $\pacc$ above a floor $\pmin$, and lower-bounds the deployment utility. This certificate must be valid under adaptive threshold selection from a finite grid of $m$ pairs on $\ncert$ samples. We give such a certificate for bounded, possibly non-monotone losses by treating the selected risk directly as a ratio rather than through a Hoeffding-style range bound. The construction couples three confidence bounds: a variance-adaptive empirical-Bernstein bound on the ratio risk, a Clopper--Pearson bound on acceptance, and a two-sided closeness bound on utility. Together they lower-bound the certified policy's utility absolutely and to within $2\gammau$ of the best over the \emph{certified set}, both non-vacuous whenever feasible; a regime-scoped third leg matches an external oracle, informative only where the risk margin $\gammar < α$ and vacuous at the headline operating points. Relative to the range-only Hoeffding-ratio construction this sharpens the acceptance-floor dependence from $1/\pmin$ to $1/\sqrt{\pmin}$, and a closed-form corollary identifies a per-pair regime in which our risk bound dominates a Hoeffding conformal risk control (Hoeffding--CRC) selective bound. Empirically, on ImageNet (three ResNets) and COCO val 2017 panoptic, the certificate opens a $+22$ pp certified-acceptance frontier over Hoeffding--CRC and is ${\approx}10{\times}$ tighter than a non-vacuous matched-valid baseline; these gains are regime-scoped, not universal, and absent on ADE20K. The certifier runs in $O(\ncert m)$ time.
Berk Hayta, Hannah Laus, Simon Mittermaier +1cs.LG eess.AS stat.ML
Real-world sensor-based learning systems require uncertainty estimation that is both reliable and computationally efficient. Evidential Deep Learning (EDL) provides single-pass uncertainty estimation by modeling the class probabilities via Dirichlet distributions, where the Dirichlet parameters are predicted by a learned neural network mapping. However, this approach can lead to computational challenges, as Dirichlet expected objectives are more complex than standard supervised learning losses, complicating their analysis and implementation. We address this issue by approximating the objective of the first-order empirical risk minimization problem induced by EDL with a plug-in loss evaluated at the Dirichlet mean and show that, under mild assumptions, the approximation error decays with growing evidence for a broad class of loss functions, including mean-squared error and cross-entropy loss. As a special case, our analysis provides justification for the use of softmax in the context of uncertainty estimation, since under a particular evidence-to-Dirichlet mapping, our framework includes the standard softmax classifier. We validate the proposed simplified objectives on the Google Speech Commands dataset and show that they achieve predictive accuracy and selective prediction performance comparable to classical EDL, while being simpler to implement using standard deep learning losses and training pipelines. To the best of our knowledge, this empirical analysis is the first to obtain coverage-accuracy trade-offs for speech recognition tasks through EDL.