Paper introduces a method to evaluate abstaining classifiers by considering missing predictions as counterfactuals.
problem Lack of a principled approach to evaluate and compare abstaining classifiers, especially when missing predictions are important.
method Develops a novel approach to treat abstentions as missing data, defining counterfactual scores and using causal inference methods to estimate them.
result Shows that under certain conditions, counterfactual scores can be identified and estimated efficiently, improving evaluation of abstaining classifiers.
Adversarial robustness improved by abstaining from decisions.
problem Improving classification accuracy in the presence of adversarial perturbations.
method Introducing an abstain option in binary classification problems, using metrics to quantify performance and robustness.
result There is a tradeoff between nominal performance and adversarial robustness.
We explore adversarial robustness in the setting in which it is acceptable for a classifier to abstain---that is, output no class---on adversarial examples. Adversarial examples are small perturbations of normal inputs to a classifier that cause the classifier to give incorrect output; they present security and safety …
A binary classifier capable of abstaining from making a label prediction has two goals in tension: minimizing errors, and avoiding abstaining unnecessarily often. In this work, we exactly characterize the best achievable tradeoff between these two goals in a general semi-supervised setting, given an ensemble of predict…
Selective classification improves trading strategies by abstaining from predictions.
problem Designing effective trading strategies using selective classification.
method Extends binary or multi-class classifiers to allow abstaining from predictions, evaluates across different feature sets and classifiers.
result Selective classifiers can improve trading performance by avoiding poor predictions.
Paper introduces SPADE method to protect classifiers from OOD and adversarial samples.
problem Protecting classifiers from out-of-distribution and adversarial samples.
method SPADE method based on GEV model in latent space.
result Provable protection against OOD and adversarial samples.
Study selective classification with limited feedback in online learning.
problem Resource-limited and safety-critical domains where feedback is only received on abstentions.
method Versioning-based schemes for minimizing mistakes and abstentions.
result Constructed schemes that make few mistakes and minimal excess abstentions.
We introduce a novel method to combat label noise when training deep neural networks for classification. We propose a loss function that permits abstention during training thereby allowing the DNN to abstain on confusing samples while continuing to learn and improve classification performance on the non-abstained sampl…
Framework to mitigate adversarial attacks by allowing classifiers to abstain.
problem Vulnerability of classifiers to adversarial examples.
method Adversarial training with a rejection option.
result ATRO framework improves classifier reliability against adversarial attacks.
This work builds a fair classification algorithm that abstains from making predictions.
problem Building a fair classification algorithm that incorporates human decision-making and avoids disparities.
method Formalizes the problem of risk minimization under fairness and abstention constraints, derives the optimal classifier, and proposes a post-processing algorithm using unlabeled data.
result The proposed algorithm achieves fairness and abstention guarantees independently of the initial classifier, provided sufficient unlabeled data is available.
This paper characterizes and designs loss functions for robust classification with abstention.
problem Ensuring robustness against adversarial attacks and knowing when to abstain from prediction.
method Proposes adversarial robust reject option loss and characterizes surrogates for calibration.
result Shifted Double Ramp Loss and Shifted Double Sigmoid Loss satisfy the calibration conditions.
We consider using an ensemble of binary classifiers for transductive prediction, when unlabeled test data are known in advance. We derive minimax optimal rules for confidence-rated prediction in this setting. By using PAC-Bayes analysis on these rules, we obtain data-dependent performance guarantees without distributio…
p-DkNN uses deep representations to detect out-of-distribution data with statistical tests.
problem Lack of reliable confidence estimates in neural networks for safety-critical applications.
method Statistical testing of deep neural network's intermediate hidden representations.
result p-DkNN enables more accurate and reliable predictions by abstaining from incorrect predictions.
Deviation-based learning improves recommender systems by abstaining from recommending choices users might follow.
problem Recommender systems learn from user choices but can stall if users blindly follow recommendations.
method The recommender learns user knowledge by observing choices, abstaining from recommending a choice when multiple alternatives produce similar payoffs.
result Learning rate and social welfare improve when the recommender abstains from recommending certain choices.
In safety-critical applications of machine learning, it is often important to abstain from making predictions on low confidence examples. Standard abstention methods tend to be focused on optimizing top-k accuracy, but in many applications, accuracy is not the metric of interest. Further, label shift (a shift in class …
New AI error correctors improve classifier performance with provable guarantees.
problem Improving AI classifier performance with scarce training data.
method Weakly supervised AI error correctors with performance guarantees.
result Provable performance guarantees for AI error correction.
Algorithm predicts with optimal loss by abstaining from uncertain test examples.
problem Predicting with training data not matching test data.
method Transductive abstention algorithm using labeled and unlabeled test examples.
result Optimal prediction loss guarantees with additional term for abstaining cost.
Unified Uncertainty Calibration improves AI predictions by combining different types of uncertainty.
problem AI classifiers struggle with uncertainty, leading to miscalibrated predictions and poor performance.
method Unified Uncertainty Calibration (U2C) combines aleatoric and epistemic uncertainties to improve prediction quality.
result U2C outperforms traditional reject-or-classify methods across various ImageNet benchmarks.
Abstaining classifiers have been widely used in cost-sensitive applications to avoid ambiguous classification and reduce the cost of misclassification. Previous abstaining classification models rely on cost information, such as a cost matrix or cost ratio. However, it is difficult to obtain or estimate costs in practic…
We deal with the \textit{selective classification} problem (supervised-learning problem with a rejection option), where we want to achieve the best performance at a certain level of coverage of the data. We transform the original m-class classification problem to (m+1)-class where the (m+1)-th class represents th…
New analysis reveals gaps in selective classifiers, guiding improvements.
problem Improving selective classifiers to match perfect-ordering oracle performance.
method Formalized selective classification gap, decomposed into five sources of looseness.
result Monotone post-hoc calibration has limited impact on closing the gap.
Efficient active learning with abstention reduces label complexity exponentially.
problem Achieving high accuracy with minimal labels.
method Developed a computationally efficient active learning algorithm with abstention.
result Achieves polylog(1/ε) label complexity, reducing by an exponential factor.
Paper tackles SCOD problem with optimal strategy and empirical validation.
problem Designing reliable prediction models abstaining from uncertain predictions.
method Bayes classifier for ID data and stochastic linear selector in 2D space.
result POSCOD method outperforms existing OOD methods.
Unified framework denoises data and abstains from uncertain predictions.
problem Data quality and predictive uncertainty in deep neural networks.
method Unified filtering framework leveraging data density.
result Framework outperforms state-of-the-art techniques in denoising and abstaining.
Algorithm learns from both labeled and arbitrary test examples, giving guarantees for bounded VC dimension classes.
problem Learning from arbitrary test examples, not just perturbations.
method Selective transductive learning algorithm that outputs abstaining predictions.
result Nontrivial guarantees for bounded VC dimension classes with arbitrary train and test distributions.
We consider the problem of n-class classification (n≥2), where the classifier can choose to abstain from making predictions at a given cost, say, a factor α of the cost of misclassification. Designing consistent algorithms for such n-class classification problems with a `reject option' is the main goal of t…
In the setting of sequential prediction of individual {0,1}-sequences with expert advice, we show that by allowing the learner to abstain from the prediction by paying a cost marginally smaller than 21 (say, 0.49), it is possible to achieve expected regret bounds that are independent of the time horizon …
A new method for selective classification trades off accuracy for coverage.
problem Selective classification allows a classifier to abstain from predicting some instances.
method Optimizes a collection of class-wise decoupled one-sided empirical risks.
result The method achieves near-optimal coverage in high target accuracy regimes.
Multi-expert L2D underfits more severely, requiring new methods.
problem Underfitting in multi-expert L2D settings.
method PiCCE (Pick the Confident and Correct Expert), a surrogate-based method.
result PiCCE effectively reduces multi-expert L2D to a single-expert-like problem, resolving underfitting.
Fusion of robustness and uncertainty techniques improves adversarial defense.
problem Adversarial attacks on deep neural networks.
method Integrating uncertainty quantification into randomized smoothing for robustness guarantees.
result Improved robustness guarantees for uncertainty aware classifiers.
New loss function helps models avoid noisy labels, improving robustness.
problem Designing robust models for datasets with noisy labels.
method Introduced a gambler's loss function that encourages models to abstain from learning noisy data points.
result Training with gambler's loss leads to improved robustness and generalization across various tasks.
Paper analyzes robustness of non-Lipschitz networks, proving powerful adversarial attacks but offering solutions.
problem Adversarial attacks on deep networks, especially non-Lipschitz networks.
method Developed an attack model that abstracts the challenge of adversarial robustness, proving the power of such attacks and offering solutions.
result Proves powerful adversarial attacks on non-Lipschitz networks but offers solutions with abstention.
Minimizes indecisions in selective classification to control misclassification rates.
problem Controlling misclassification rates in high-risk scenarios.
method Using indecisions to control misclassification rates, even below Bayes optimal.
result Control of misclassification rates to any user-specified level, even below Bayes optimal.
Study improves resilience against adversarial clean-label attacks in real and noisy settings.
problem Ensuring accurate predictions in the presence of adversarial clean-label samples.
method Sequential learning from a stream of i.i.d. data, allowing abstention for uncertain predictions.
result Theoretical analysis and adaptations for the agnostic setting with a clean-label adversary and noise.
A new policy learning method allows policies to abstain when uncertain, improving safety and applicability.
problem Risk of making decisions without full confidence in uncertain predictions.
method Policy learning with abstention, identifying near-optimal policies and constructing an abstention rule.
result Improved safety and applicability in policy learning, with theoretical guarantees.
Optimal selective classification using likelihood ratios improves model reliability.
problem Enhancing predictive model reliability by allowing uncertain predictions.
method Neyman--Pearson lemma applied to likelihood ratios for optimal selection.
result Neyman--Pearson-informed methods outperform existing baselines under covariate shifts.
Adaptive algorithm identifies best arm with abstention, showing phase transition from polynomial to exponential error probability.
problem Bayesian best-arm identification with abstention to reduce undetected error.
method Adaptive algorithm PGWS that optimally uses abstention budget.
result Introducing any positive abstention budget induces an exponential decay in undetected error probability.
A new model for sequential prediction handles adversarial examples by allowing abstention.
problem Sequential prediction algorithms fail with adversarial examples, leading to incorrect predictions.
method Proposes a new model that allows abstention from predictions on adversarial examples, scaling error with VC dimension.
result A learner's error scales with the VC dimension of the hypothesis class, matching the stochastic setting.
Approves updates to machine learning models in healthcare based on accumulating data.
problem Designing policies to autonomously approve updates to machine learning algorithms in non-stationary settings.
method Learning-to-approve (L2A) approach that uses accumulating monitoring data to learn how to approve modifications.
result L2A learns to abstain when performance drops are common and approves beneficial modifications quickly when the distribution is stable.
Algorithm learns binary function efficiently under arbitrary covariate shift.
problem Learning binary function under arbitrary distributions P and Q.
method PQ-learning algorithm using reliable learner with selective classification.
result Polynomial-time algorithm for covariate shift learning.
We study active learning where the labeler can not only return incorrect labels but also abstain from labeling. We consider different noise and abstention conditions of the labeler. We propose an algorithm which utilizes abstention responses, and analyze its statistical consistency and query complexity under fairly nat…
New approach to robustly reliable learners against instance-targeted attacks.
problem Addressing data poisoning attacks on robustly reliable learners.
method Defining regularized robustly reliable learners and efficient algorithms.
result Efficient algorithms for robustly reliable learners with sublinear runtime.
FactTest assesses LLM factuality with Type I error control.
problem Lack of rigorous factuality verification for LLMs.
method Formulates factuality testing as hypothesis testing, ensuring Type I and II error control.
result Improves model accuracy by over 40% in abstaining from unknown questions.
Learning compact and interpretable representations is a very natural task, which has not been solved satisfactorily even for simple binary datasets. In this paper, we review various ways of composing experts for binary data and argue that competitive forms of interaction are best suited to learn low-dimensional represe…
This thesis enhances ML reliability by selectively abstaining from predictions when uncertain.
problem Improving reliability in machine learning systems, especially in high-stakes domains.
method Exploiting uncertainty signals from training trajectories to develop lightweight, post-hoc abstention methods compatible with differential privacy.
result A robust trajectory-based approach to selective prediction that maintains high accuracy under privacy noise.
In contrast to conventional (single-label) classification, the setting of multilabel classification (MLC) allows an instance to belong to several classes simultaneously. Thus, instead of selecting a single class label, predictions take the form of a subset of all labels. In this paper, we study an extension of the sett…
We study pool-based active learning with abstention feedbacks where a labeler can abstain from labeling a queried example with some unknown abstention rate. This is an important problem with many useful applications. We take a Bayesian approach to the problem and develop two new greedy algorithms that learn both the cl…
We study pool-based active learning with abstention feedbacks, where a labeler can abstain from labeling a queried example with some unknown abstention rate. This is an important problem with many useful applications. We take a Bayesian approach to the problem and develop two new greedy algorithms that learn both the c…