This special issue of Mathematical Foundations of Computing presents a curated selection of seven high-quality papers originally presented at the 6th Machine Learning and Optimization (MLO 2023) conference. The collection highlights the critical synergy between rigorous mathematical theory and evolving practical applications, spanning fields from financial risk management to multimodal artificial intelligence. Out of 60 submissions, only 7 were selected for this issue, representing an acceptance rate of approximately 11.6% and ensuring a high standard of academic rigor.
Theoretical Foundations: Graph Theory and Submodularity
A significant portion of this issue is dedicated to advancing the theoretical understanding of discrete structures. In the realm of graph theory, researchers have provided a definitive characterization of symmetric factorizations in complete graphs. Specifically, the work proves that for $n \ge 4$, the complete graph $K_n$ possesses a symmetric 1-factorization if and only if $n$ is a power of 2 or belongs to the set $\{6, 12\}$. This result clarifies long-standing questions regarding structural symmetry and homogeneity in network science.
Furthermore, the issue explores the complexities of submodular maximization. Authors developed new greedy and threshold descent algorithms to maximize $k$-submodular functions under $p$-system constraints. A key contribution of this research is the establishment of provable approximation guarantees: a ratio of $1/(1+p)$ for monotone functions and $1/(2+p)$ for non-monotone cases. The introduction of threshold descent algorithms demonstrates a strategic trade-off, significantly reducing computational time complexity while maintaining high approximation accuracy.
Robust Optimization and Financial Modeling
The application of optimization to financial markets is a recurring theme. Two featured papers address the challenge of portfolio selection under uncertainty using distributionally robust optimization. By employing the Wasserstein distance and Conditional Value-at-Risk (CVaR), these models account for "worst-case" scenarios where the underlying statistical distributions of market data may deviate from observed patterns. This approach ensures that investment portfolios remain resilient even when traditional statistical assumptions fail, providing a more reliable framework for real-world financial management.
Advancements in Signal Processing and Multimodal AI
In the field of signal processing, the special issue highlights improvements in compressed sensing. By adapting the "Heavy-ball" method—a technique that incorporates momentum into gradient descent—researchers have enhanced the performance of thresholding algorithms. This acceleration allows for faster and more accurate reconstruction of signals from sparse data, a critical requirement in modern telecommunications and imaging.
The collection also ventures into multimodal AI with a focus on sentiment analysis. The proposed MSA-HCL model utilizes hybrid contrastive learning to bridge the gap between different data types, such as text and images. By aligning features across modalities (inter-modal) while simultaneously preserving the unique context within each individual modality (intra-modal), the model achieves a more nuanced understanding of sentiment than traditional single-modality approaches.
Algorithmic Innovations in Network Science
Finally, the issue addresses the scalability of network analysis. New algorithms are proposed for the decomposition and maintenance of $k$-hypercores within massive, edge-weighted hypergraphs. These methods are essential for managing the increasing complexity of modern datasets, where relationships are often higher-order and weighted by importance or frequency.
In summary, this special issue serves as a testament to the vibrant intersection of machine learning and mathematical optimization. By grounding practical innovations in robust theoretical frameworks, these papers contribute significantly to the ongoing evolution of computational science.
Summary updated