This survey highlights how optimization- and certification-oriented reasoning can provide a useful framework for reasoning about differences in models, supporting tasks ranging from model training and selection to auditing and certification.
Abstract
Modern machine learning (ML) increasingly relies on complex models whose behavior is difficult to characterize beyond empirical performance metrics. Across a wide range of tasks, including prediction, generation, and decision-making, models with similar empirical performance can exhibit markedly different properties in terms of their transparency, interpretability, robustness, fairness, privacy, and certifiability. This survey highlights how optimization- and certification-oriented reasoning can provide a useful framework for reasoning about such differences, supporting tasks ranging from model training and selection to auditing and certification. We review and synthesize recent advances at the intersection of combinatorial optimization (CO) and trustworthy ML, covering both training and post-training tasks, including interpretable model learning, explanation generation, robustness analysis, fairness auditing, model compression, and privacy attacks and protections. Across these domains, CO formulations offer additional capabilities over purely heuristic approaches, e.g., gradient-based ones, notably global guarantees, formal certificates, and explicit treatment of trade-offs. While scalability remains an important challenge, continued progress in solvers and hybrid algorithms suggests a growing role for CO in the design and deployment of trustworthy ML systems.
Assessing the quality of Large Language Model (LLM) outputs becomes especially challenging in high-branching settings, where a single prompt yields many plausible candidates. Existing verifiers typically operate on the surface text (e.g., reward models, LLM judges, majority voting) or on confidence proxies derived from token probabilities, both of which can be brittle: the former can be influenced by stylistic artifacts, while the latter is often miscalibrated. In this paper, we study a third source of information—the model’s hidden states—for binary correctness verification in tasks with a reliable success/failure signal (e.g., deterministic checkers or reference-grounded answers). We find that correct and incorrect solutions exhibit measurable geometric differences in their hidden-state trajectories. To isolate this signal with minimal modeling assumptions, we introduce C LUE (Clustering and Experience-based Verification) , a training-free, non-parametric verifier. C LUE summarizes each reasoning trace by an activation delta —the difference between hidden states at the start and end of the explicit reasoning span— and predicts correctness by comparing this delta to two class centroids computed from labeled experience. Across math (AIME 24/25), scientific QA (GPQA), and a multi-domain benchmark (WebInstruct-verified), C LUE improves selection and reranking
Zhenwen Liang, Ruosen Li, Yujun Zhou et al.· Annual Meeting of the Associ...· 0 citations
Privacy regulations and institutional policies limit the sharing of educational data, constraining reproducibility in learning analytics. Prior evaluations of synthetic data on benchmarks such as OULAD have examined statistical or adversarial synthesizers in isolation, rarely jointly assessing fidelity, utility, privacy, and explainability. We compare three synthesis paradigms, statistical (Gaussian Copula), adversarial (CTGAN), and diffusion-based (TabDDPM), on two benchmarks (OULAD: 32,593 records; ASSISTments: 8,519) across five evaluation axes: distributional fidelity (SDMetrics), downstream utility (Train on Synthetic, Test on Real), discriminative realism (classifier two-sample test), membership-inference privacy, and feature-importance preservation (SHAP). The pipeline is repeated over five random seeds with bootstrap confidence intervals and Bonferroni-corrected permutation tests (<inline-formula> <tex-math notation="LaTeX">$\alpha \prime \approx ~0.0028$ </tex-math></inline-formula>). Four findings emerge: First, TabDDPM delivers the strongest classification utility: on OULAD, a Random Forest achieves TSTR AUC <inline-formula> <tex-math notation="LaTeX">$= 0.962~\pm ~0.001$ </tex-math></inline-formula>, within 0.5 percentage points of the real-data baseline. Second, all synthesizers exhibit near-chance membership-inference risk under the evaluated kNN-based threat model (worst-case effective AUC <inline-formula> <tex-math notation="LaTeX">$\le 0.527$ </tex-math></inline-formula>). Third, distributional fidelity does not predict task utility; CTGAN scores highest on SDMetrics yet does not yield the smallest utility gap. Fourth, TabDDPM best preserves real-data feature-importance rankings on OULAD (Spearman <inline-formula> <tex-math notation="LaTeX">$\rho =0.846$ </tex-math></inline-formula>, p < 0.001). ASSISTments SHAP correlations (<inline-formula> <tex-math notation="LaTeX">$\rho ~ \ge 0.950$ </tex-math></inline-formula>) reflect a low-dimensionality ceiling effect rather than meaningful synthesis quality differences. These results provide task-driven guidance for selecting a synthesizer in learning analytics. Scope is limited to static tabular benchmarks; temporal, sequential, multimodal, and fairness-aware synthesis remain outside the present scope. Runtime results are based on CPU execution, so neural synthesizers may run faster under GPU acceleration.
Design campaigns in chemistry, materials science, and machine learning share a bottleneck: determining how good a candidate truly is requires an expensive evaluation - an experiment, a first-principles simulation, or a full training run. Machine-learning surrogates that predict these outcomes are increasingly used not only to propose candidates but to grade them, and even to feed their own predictions back into the search as though they were measurements. Through mathematical analysis validated on three exhaustively ground-truthed design tasks, we establish when this practice is safe, what any certificate of safety must cost, and when the substitution provably pays. Predictive accuracy cannot anchor trust: near-perfect R^2 is compatible with worst-possible selections, and screening N candidates inflates the over-prediction at the selected candidate by a quantifiable"selection tax"with matching upper and lower bounds. Safety follows instead from an architectural rule - predictions may propose and train without restriction, but every certified conclusion must rest on true evaluations - which is sufficient with no assumptions on the surrogate, and necessary, since admitting predictions into certification with the standing of measurements opens a deterministic self-confirmation failure mode. We derive the minimal criterion under which a model may act as an oracle (rank preservation, not accuracy), show that trust must be purchased through selection-aware audits that are optimal in query complexity, and prove a dichotomy fixing when audited surrogates cut certified evaluation cost. Across 432 surrogate fits over six task-regime conditions, the audit statistic tracks deployed search performance at Spearman rank correlation 0.80-0.99, while the rank correlation of R^2 with deployed regret falls as low as 0.33; audited screening reduces certified oracle cost by a measured factor of 25.
This paper presents a structured and quantitative framework for evaluating the explainability of machine learning (ML) models trained on synthetic access control datasets. Explainability is assessed using four complementary metrics - stability, consistency, fidelity, and monotonicity - along with a normalized composite explainability score $E_{\text{norm}}$ that enables balanced comparison across models and datasets by addressing scale incompatibility among metrics. The framework is evaluated across four ML models (Logistic Regression, k-Nearest Neighbors, Naive Bayes, and a Neural Network) trained on six synthetic datasets ranging from rule-based and probabilistic to random and adversarial generation strategies. Results show that simpler models consistently produce more stable, consistent, and monotonic explanations on structured synthetic data, while neural networks exhibit greater variability and over-reliance on dominant features. Structured data generation methods were found to support stronger explainability than unstructured or adversarial approaches. Together, these findings provide practical guidance for selecting model-dataset combinations that support transparent and auditable decision-making in physical access control systems.
Ahmad N. Mohamad Jembari, Mustafa Abdallah· 2026 6th International Confe...· 0 citations
Adversarial machine learning (AML), reinforcement learning (RL), and explainable artificial intelligence (XAI) are increasingly studied as separate problems, yet their interactions under realistic threat conditions remain poorly understood. This review addresses that gap through a systematic analysis of 207 studies selected from 4447 records following the PRISMA 2020 guidelines, covering work published between 2020 and 2026 across cybersecurity and computer vision. A taxonomy of adversarial attacks is constructed across training and inference phases, defense mechanisms are examined with attention to their documented failure modes, and robustness evaluation practices are assessed across the surveyed literature. RL is analyzed in both offensive and defensive roles. Attack agents using RL achieve evasion rates of 74–97% against ML-based detectors, while RL-based defenses report robustness gains of up to 3× over static baselines under comparable threat conditions. XAI receives particular attention because the field treats it almost exclusively as a transparency mechanism, whereas the reviewed evidence shows that it also functions as an attack surface. Attribution methods such as LIME, SHAP, and Grad-CAM produce unreliable explanations under adversarial perturbation, and no system in the reviewed literature certifies that attribution properties are maintained when inputs are manipulated. The review concludes with an analysis of open problems and research directions for building systems that are robust against adaptive adversaries, interpretable under operational constraints, and auditable in environments where AI accountability is a legal requirement.
As Artificial Intelligence systems are increasingly deployed in safety-critical domains such as healthcare and finance, ensuring their trustworthiness and compliance is paramount. While Deep Neural Networks have received significant attention in formal verification, traditional models such as Decision Trees, often preferred for their interpretability, cannot inherently enforce constraints after training for fairness and stability. This paper presents a novel, comprehensive approach for the formal verification of Decision Tree classifiers using Satisfiability Modulo Theories (SMT). We propose a robust translation scheme that converts trained decision trees into logical constraints, enabling constraint inference that guarantees demographic parity and local robustness at prediction time. We implement this framework by using the z3 SMT solver and validate it on widely recognized fairness benchmarks, including UCI Adult, German Credit, and Loan Approval datasets. Experimental results demonstrate that our constrained model effectively eliminates demographic parity violations with a marginal accuracy trade-off of less than 0.2%. This approach transforms the SMT solver from a simple diagnostic tool into a provably fair inference engine suitable for regulated industries.
A. Beishekeyev, T. Umarov· Herald of Kazakh-British tec...· 0 citations