information-gain analysis

Evaluating learning-augmented online algorithms under uncertainty remains challenging, as conventional metrics focus narrowly on worst-case prediction errors, neglecting both prediction accuracy and risk sensitivity. Method: We propose a dual-track evaluation framework grounded in decision theory, jointly incorporating distance-based prediction error quantification (deterministic aspect) and risk-sensitive modeling (stochastic aspect). By embedding decision-theoretic loss functions into online algorithm analysis, we integrate prediction error modeling with risk-controllable optimization, designing novel learning-augmented algorithms for contract scheduling and 1-max search. Contribution/Results: Our approach achieves provable robustness to prediction errors, performance guarantees with tight bounds, and explicit risk controllability. It is the first to unify prediction accuracy, worst-case robustness, and risk preference within a single theoretical framework—establishing a systematic evaluation paradigm and design principle for learning-augmented online algorithms.

Yannakakis’ algorithm exhibits unstable performance and poor adaptability in query optimization, particularly under dynamic workloads. Method: This paper pioneers a machine learning–driven approach to optimizer decision-making by framing the choice of whether to apply Yannakakis’ algorithm as a binary classification task. Leveraging structural features of queries, statistical metadata, and cost model estimates, it employs supervised learning models—specifically XGBoost and Random Forest—to enable query-level adaptive selection. Contribution/Results: Unlike conventional static heuristics or hard-coded rules, the proposed method is portable across database management systems. Extensive experiments across multiple benchmarks and diverse DBMSs demonstrate an average 23.7% reduction in query latency (p < 0.01), confirming both the effectiveness and generalizability of ML-driven optimization policy decisions.

This study addresses the quantification of discriminative power in query-document relevance judgments (qrels) for information retrieval (IR) evaluation, with particular emphasis on the historically underexamined Type II error (false negatives) and its joint analysis with Type I error (false positives). Method: We systematically introduce Type II error modeling into IR evaluation for the first time and propose replacing conventional significance testing with balanced classification metrics—such as balanced accuracy—as a more principled basis for assessing qrels’ discriminative capability. A unified, comparable measurement framework is thereby established. Results: Empirical hypothesis testing and statistical significance analysis across multiple qrels generation strategies demonstrate that jointly evaluating both error types exposes latent quality deficiencies in qrels more comprehensively than traditional approaches. Balanced classification metrics robustly aggregate discriminative performance, substantially enhancing the reliability and interpretability of IR system evaluation.

This work addresses the widespread misuse of information-theoretic measures in contemporary AI practice, which often stems from overlooking estimator assumptions, failure modes, and conditions required for reliable inference. The paper introduces the first unified decision framework that systematically integrates established measures—such as entropy and mutual information—with emerging ones like integrated information (Φ) and effective information. For each measure, the framework explicitly clarifies three core considerations: suitable AI application scenarios, appropriate estimators for given data types and dimensionalities, and common pitfalls leading to misinterpretation. Standardization and operationalization of measure selection are achieved through a combination of flowcharts, a primary decision table, and Bridge Box cognitive mapping techniques. The efficacy of this framework is empirically validated across three representative tasks: representation learning, temporal influence analysis, and agent complexity assessment.

This work addresses the problem of automatically selecting optimal execution plans for general conjunctive queries (CQs) based on database statistics. It proposes the PANDA framework, which, for the first time, directly integrates information-theoretically derived tight upper bounds on intermediate relation cardinalities into the query plan generation and optimization process. The approach unifies treatment across diverse query scenarios and matches or surpasses the performance of specialized algorithms on several classical problems—including those relying on fast matrix multiplication—demonstrating both generality and efficiency. The key contribution lies in establishing a novel connection among information theory, constraint satisfaction problems, and database query optimization, thereby achieving a synergistic improvement between theoretical guarantees and practical performance.

This work addresses the misalignment between offline evaluation metrics and online performance objectives in industrial applications by establishing a unified theoretical framework that systematically quantifies the relationships among diverse evaluation metrics for the first time. By introducing the concepts of Bayes-optimal sets and regret transfer mechanisms, the study reveals structural asymmetries among metrics and provides a principled classification and relational modeling of metrics with varying mathematical forms. Theoretically characterizing metric consistency and transferability, this research offers novel insights and a methodological foundation for designing offline evaluation systems that are aligned with online objectives and backed by rigorous theoretical guarantees.

This work addresses the problem of selectivity estimation for linear queries—such as point and range queries—in dynamic databases. It introduces, for the first time, online learning theory to this setting, proposing an online estimation method based on histogram models and standard loss functions. The approach effectively adapts to time-varying data distributions and query workloads, delivering provably low regret in both static and dynamic environments. The core contribution lies in establishing tight upper and lower bounds on regret specifically for histogram-based linear queries, thereby providing the first formal online learning framework for dynamic selectivity estimation with theoretical performance guarantees.

This work addresses the problem of efficiently identifying the key training samples underlying a model’s predictions to enhance interpretability and safety. It formulates data attribution as a Bayesian information-theoretic problem, using the increase in predictive entropy—i.e., information loss—induced by removing a sample as the attribution criterion, thereby prioritizing the reduction of prediction uncertainty over fitting label noise. The approach leverages Gaussian process surrogates and tangent features for efficient approximation and introduces a scalable information gain objective coupled with a variance correction mechanism, enabling compatibility with large-scale vector database retrieval. Empirically, the method demonstrates strong performance across counterfactual sensitivity, ground-truth attribution retrieval, and coreset selection tasks, offering both theoretical rigor and scalability to modern deep architectures.

This work addresses the challenge of ambiguous user preferences during early-stage interactions on e-commerce platforms, which often leads to query ambiguity, redundant interactions, or premature convergence in recommender systems. To mitigate this, the authors propose an entropy-based Interactive Decision Support System (IDSS) that leverages entropy as a unified signal to dynamically maintain candidate sets and quantify attribute-level uncertainty. The system actively elicits user feedback through questions selected to maximize information gain, thereby clarifying preferences, and subsequently incorporates residual uncertainty into the recommendation phase for ranking and diversified presentation. Experimental results on a simulation environment built from real user reviews demonstrate that the approach significantly reduces ineffective interactions while enhancing recommendation diversity, informativeness, and overall quality, thereby improving system transparency and users’ sense of control.