Fairness in machine learning has become increasingly crucial, particularly in graph-based models where biased representations can reinforce societal inequalities. Traditional fairness-aware learning methods on graphs focus on graph rewiring, debiasing node embeddings, adversarial learning, and additional fairness constraints. However, these approaches often struggle to balance fairness and task performance. We propose a novel edge partitioning strategy that creates two distinct subgraphs, maintaining a balance between bias and diversity. We categorize edges as homophilic or heterophilic depending on the sensitive attribute of the corresponding node pairs. An edge is s-homophilic if it joins two nodes with the same sensitivity value, otherwise s-heterophilic. The partition splits the input graph into two subgraphs, both containing all nodes, one with only s-homophilic edges and the other with s-heterophilic ones. Using a Graph Neural Network (GNN), we obtain independent node representations from both graphs, which are then aggregated into a unified node embedding. To enforce fairness, we jointly optimize a primary task loss and a fairness loss, ensuring predictive accuracy and bias mitigation. We evaluate our approach on three benchmark datasets and find that it achieves improved fairness metrics while maintaining accuracy comparable to that of existing state-of-the-art methods.

Random mechanisms including mutations are an internal part of evolutionary algorithms, which are based on the fundamental ideas of Darwin’s theory of evolution as well as Mendel’s theory of genetic heritage. In this paper, we debate whether pseudo-random processes are needed for evolutionary algorithms or whether deterministic chaos, which is not a random process, can be suitably used instead. Specifically, we compare the performance of 10 evolutionary algorithms driven by chaotic dynamics and pseudo-random number generators using chaotic processes as a comparative study. In this study, the logistic equation is employed for generating periodical sequences of different lengths, which are used in evolutionary algorithms instead of randomness. We suggest that, instead of pseudo-random number generators, a specific class of deterministic processes (based on deterministic chaos) can be used to improve the performance of evolutionary algorithms. Finally, based on our findings, we propose new research questions.

A large amount of research effort has been dedicated to adapting boosting for imbalanced classification. However, boosting methods are yet to be satisfactorily immune to class imbalance, especially for multi-class problems. This is because most of the existing solutions for handling class imbalance rely on expensive cost set tuning for determining the proper level of compensation. We show that the assignment of weights to the component classifiers of a boosted ensemble can be thought of as a game ofTug of Warbetween the classes in the margin space. We then demonstrate how this insight can be used to attain a good compromise between the rare and abundant classes without having to resort to cost set tuning, which has long been the norm for imbalanced classification. The solution is based on a lexicographic linear programming framework which requires two stages. Initially, class-specific component weight combinations are found so as to minimize a hinge loss individually for each of the classes. Subsequently, the final component weights are assigned so that the maximum deviation from the class-specific minimum loss values (obtained in the previous stage) is minimized. Hence, the proposal is not only restricted to two-class situations, but is also readily applicable to multi-class problems. Additionally, we also derive the dual formulation corresponding to the proposed framework. Experiments conducted on artificial and real-world imbalanced datasets as well as on challenging applications such as hyperspectral image classification and ImageNet classification establish the efficacy of the proposal.