Multiobjective Optimization, Clustering and AI: Towards Intelligent Discovery

Modern scientific discovery increasingly relies on computational methods that can explore complex search spaces, identify hidden structures in data, and learn meaningful representations from large and heterogeneous datasets. This tutorial talk will discuss ways of addressing these tasks from the perspectives of multiobjective optimization, clustering, and artificial intelligence. This talk will present the fundamental concepts, algorithms, and emerging connections among these three areas, with an emphasis on their role in intelligent discovery. The tutorial will begin with the principles of multiobjective optimization, including conflicting objectives, Pareto dominance, Pareto-optimal solutions, and trade-off analysis. Two influential evolutionary approaches, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and the Archived MultiobjectiveSimulated Annealing (AMOSA) algorithm, will be introduced to illustrate population-based and simulated-annealing-based strategies for exploring multiobjective search spaces. An application to rational drug design will be briefly demonstrated. The second part will introduce clustering as an unsupervised learning paradigm for discovering structure in unlabeled data. Fundamental methods, including K-means and Fuzzy C-means (FCM), will be explained, highlighting their assumptions, optimization principles, strengths, and limitations. Building on these foundations, the tutorial will examine multiobjective optimization for clustering, where competing criteria such as cluster compactness and separation can be optimized simultaneously. An innovative technique for providing a single comprehensive clustering of the data utilizing a novel combination of the multiple Pareto-optimal solutions in a supervised framework will be discussed. An application to clustering gene expression data will be briefly demonstrated. The final part will connect these classical computational intelligence techniques with modern AI. Topics will include some historical facts about the evolution from traditional machine learning to deep learning, representation learning, attention mechanisms, and Transformer architecture. Selected contemporary trends, including foundation models, generative AI, retrieval-augmented generation, and AI-assisted scientific discovery, will be briefly mentioned alongside their opportunities and limitations. It is expected that on conclusion of the tutorial, the participants will have a fair grasp of metaheuristics for multiobjective optimization, a couple of widely used clustering algorithms, a technique for multiobjectiveclustering, and the transformer model and some of its developments.