This research introduces a novel deep learning based oversampling method for the class imbalance problem. Compared to previous methods, our approach defines the oversampling process as a composition of multiple decisions. This allows the deep learning classifier to learn the optimal mechanism for each decision from the ground truth data patterns, enabling more fine-grained control over the data oversampling process. We provide experiments on real-world datasets to demonstrate the superiority of our solution over the state-of-the-art oversampling methods.
11 April 2025 in Proceedings of the AAAI Conference on Artificial Intelligence
Influence maximization is a key topic in data mining, with broad applications in social network analysis and viral marketing. In recent years, researchers have increasingly turned to machine learning techniques to address this problem. By learning the underlying diffusion processes from data, these methods improve the generalizability of solutions while optimizing objectives to identify the optimal seed set for maximizing influence. Nonetheless, two fundamental challenges remain unresolved: (1) While Graph Neural Networks (GNNs) are increasingly employed to learn diffusion models, their traditional architectures often fail to capture the complex dynamics of influence diffusion, (2) Designing optimization objectives is inherently difficult due to the combinatorial explosion associated with solving this problem. To address these challenges, we propose a novel framework, DeepSN. Our framework employs sheaf neural diffusion to learn diverse influence patterns in a data-driven, end-to-end manner, providing enhanced separability in capturing diffusion characteristics. We also propose an optimization technique that accounts for overlapping influence between vertices, significantly reducing the search space and facilitating the identification of the optimal seed set efficiently. Finally, we conduct extensive experiments on both synthetic and real-world datasets to demonstrate the effectiveness of our framework.
Solving a Math word problem is complex due to the existence of different variations branching out as linear, quadratic, etc. as well as ratios, percentages, fractions, etc. Among the existing approaches to solve Math word problems, template based machine learning approach is found to be the most prominent. However, existing template based methods have a major limitation where they have to work with the high dimensionality of the training feature space. In this research, we focus on improving these template based techniques by reducing the dimensionality of the feature space used for classification. A two-phase classifier is trained using the templates for simultaneous linear equations to predict the correct equation set and the final answer. This new system was able to achieve an accuracy of 91.2% on a state-of-the-art data set, which outperforms the current highest accuracy of 79.7% reported on the same data set.
Graph Neural Networks (GNNs) have become the dominant paradigm for molecular representation learning. However, traditional GNNs have their representation capacities fundamentally limited by the 1-dimensional Weisfeiler-Lehman (1-WL) test, preventing them from distinguishing structurally different molecules. To address this challenge, we propose a domain-specific curvature augmentation that captures multi-scale and domain-specific structural information around nodes and edges in molecular graphs. Our approach introduces the Augmented Curvature Graph Neural Network (AC-GNN), a simple neural plugin that seamlessly integrates curvature information into any existing GNN architecture by modifying the adjacency matrix with curvature-weighted edges. We provide theoretical guarantees demonstrating that AC-GNN achieves strictly greater expressive power than 1-WL bounded methods, enabling the differentiation of molecular graphs that appear identical to traditional GNNs. Extensive experiments on benchmark datasets for molecular classification and regression tasks demonstrate consistent and significant performance improvements of AC-GNN across multiple GNN backbones.