false positive mitigation

This paper investigates the optimal individual behavior classification problem under outcome performativity—where classifier outputs causally influence the behavior of classified agents. We formalize the decision-making process via a strategic response model and conduct rigorous theoretical analysis. Our main contribution is the first formal proof that the optimal classifier must be either a standard threshold rule or a *negative-threshold rule*, which assigns positive outcomes to agents with *lower* predicted propensity—a counterintuitive structure that strictly dominates conventional thresholding under performative settings. We establish a structural optimality theorem, constructively characterize the mechanism by which negative-threshold rules achieve superior accuracy, and generalize the result to arbitrary objective functions, including weighted misclassification losses. This finding challenges the intuitive “high-propensity-first” allocation principle and provides a provably optimal theoretical foundation and actionable design guidelines for deploying classifiers in performative environments.

This study addresses the critical issue that existing selective prediction methods in signal domains—such as anomalous sound detection and AI-generated image forensics—often yield a false sense of security due to the use of uncalibrated thresholds, resulting in actual error rates that substantially exceed users’ prescribed risk budgets. The work presents the first systematic audit of four distribution-free calibration rules (NAIVE, Hoeffding, Clopper–Pearson, and Betting) regarding their risk control performance on both real and synthetic data. Findings reveal that NAIVE exceeds the risk budget in 49–73% of experiments; Clopper–Pearson and Betting achieve zero violations under exchangeability but suffer 9–30% violation rates when deployed in grouped settings where exchangeability fails. Group-wise thresholding restores valid risk control at the cost of reduced coverage. The study underscores the pivotal role of tight confidence bounds for effective coverage and identifies uncalibrated thresholds as the root cause of risk miscontrol.

This study investigates the impact of label noise on model abstention behavior in out-of-distribution (OOD) classification. We propose BCOPS, a conformal prediction-based algorithm that— for the first time under label noise—constructs prediction sets with statistically guaranteed coverage, and systematically evaluates its abstention rate and robustness on unseen classes during training. Experiments on synthetic data and real-world benchmarks (e.g., CIFAR-10/CIFAR-100) reveal that even low-level label noise (≤5%) substantially increases OOD abstention rates, undermining model reliability. Our key contributions are threefold: (i) uncovering an implicit interference mechanism by which label noise distorts OOD abstention; (ii) empirically characterizing the degradation pattern of BCOPS’s coverage fidelity under noise; and (iii) establishing a novel evaluation paradigm for noise-robust trustworthy anomaly detection.

This work addresses the trade-off between human annotation cost and system accuracy in human-in-the-loop classification by proposing an optimization framework based on a dual-threshold strategy. By setting upper and lower confidence thresholds, the system automatically processes high-certainty samples while routing only ambiguous cases to human reviewers. The approach formalizes the human-AI collaboration problem, identifies the critical region where human intervention yields diminishing returns, and quantifies the marginal benefit of manual review across diverse scenarios through probabilistic score modeling, Monte Carlo simulation, and optimization algorithms. Empirical evaluations demonstrate the framework’s generality and effectiveness across multiple domains—including entity resolution, fraud detection, medical triage, and content moderation—achieving high accuracy while substantially reducing human workload.

In industrial quality inspection, anomaly detection suffers from poor robustness due to high noise levels and sparse defective samples. To address this, we propose Iterative Refinement of Pseudo-labels (IRP), a self-supervised method that alternately evaluates sample credibility and removes misleading instances under feature-space consistency constraints—effectively purifying the training set dynamically without human annotations and generating high-fidelity self-supervised signals. IRP introduces the novel paradigm of “iterative data refinement,” significantly enhancing model robustness against label noise and cross-domain generalization capability. Evaluated on KSDD2 and MVTec AD benchmarks, IRP consistently outperforms existing unsupervised and self-supervised methods. Notably, under high-noise conditions, it achieves substantial improvements in detection accuracy and reduces false positive rates by over 25%.

This work proposes a novel representation learning framework that addresses the limited representational capacity of existing methods in complex scenes by integrating adaptive multi-scale fusion with contrastive learning. The approach dynamically aggregates multi-level features and incorporates a structure-aware contrastive loss, thereby enhancing the model’s ability to jointly capture fine-grained semantics and global contextual information. Extensive experiments demonstrate that the proposed framework consistently outperforms state-of-the-art methods across multiple benchmark datasets, achieving substantial improvements in both accuracy and robustness. These results establish a promising new direction for unsupervised and semi-supervised representation learning.

This work proposes a sequential testing–based early-stopping strategy for binary ensemble classifiers to reduce inference overhead while strictly bounding the divergence rate from predictions of the full ensemble. The approach terminates evaluation as soon as a decisive majority emerges during the sequential assessment of base models. Under three optimality criteria, the strategy can be formulated as a linear programming problem, enabling efficient computation of the optimal stopping rule. Experimental results on UCI and Grinsztajn benchmark datasets demonstrate that the method achieves an average speedup exceeding 4× while consistently maintaining prediction divergence below 0.1%.

This work addresses the misalignment between confidence signals from machine learning–based detection models in Security Operations Centers (SOCs) and analysts’ decision-making objectives, which often leads to overlooked false positive costs, increased false negative risks, and alert overload. To bridge this gap, the authors propose a decision-aware trust signal alignment framework that, for the first time, incorporates asymmetric decision costs into trust signal design. By leveraging posterior calibration, lightweight uncertainty indicators, and cost-sensitive thresholds, the approach achieves model-agnostic decision alignment without modifying the underlying detection model. Experiments on the UNSW-NB15 dataset demonstrate that the method reduces weighted loss by several orders of magnitude, substantially decreases false negatives, and lays the groundwork for future human-in-the-loop cybersecurity research.

This work addresses the challenge of error localization in autonomous analytical agents, which often lack unsupervised auditing mechanisms when failures occur during end-to-end data analysis. The authors propose a novel anomaly detection method that requires no error annotations by modeling normal analytical behavior and assigning anomaly scores to deviant operations. Key contributions include elucidating the relationship between error localization and scoring strategies, introducing a reconstruction-length-based approach to quantify error propagation, establishing a false positive control mechanism relying solely on exchangeability assumptions, and deriving the first theoretical lower bound on error identifiability. Both theoretical analysis and experiments demonstrate that single errors can be flagged either locally or through diffusion, false positive rates remain controllable, and identification capability is primarily constrained by representation dimensionality rather than training data volume.

This work addresses the high-risk, non-human misclassifications arising from machine learning models in safety-critical applications by proposing a post-hoc correction framework based on two gradient-boosted decision tree (GBDT) classifiers, without requiring retraining of the primary model. The method identifies high-risk errors and applies conservative corrections, achieving substantial improvements in system safety while incurring minimal inference overhead (<1.85%). Evaluated on the ISIC and SICAPv2 datasets, the approach reduces high-risk errors by 34.1% and 12.57%, respectively, and attains superclass diagnostic safety rates of 90.41% and 92.13%, significantly outperforming the baseline based on maximum class probability.