Bayesian optimization (BO) is a class of methods with state-of-the-art performance that deal with black-box functions, where such a function is defined as having unknown gradients (hence not being able to apply classical optimization), being corrupted by noise and being very expensive to evaluate. We can find some example while performing hyper-parameter tuning of deep learning algorithms with respect to some loss function on a particular dataset, in performing hyper-parameter tuning of deep reinforcement learning algorithms but also on financial portfolio optimization, energy and more. Some scenarios require the simulatenous optimization of inversely correlated black-boxes, as optimizing both the predictive error and predictive time of a deep neural network. In this scenario, named multi-objective BO, we are interested to find the optimal Pareto set of solutions. We can also take into account several black-box constraints and evaluations in parallel, which is the specific topic of this seminar, where we present two information theoretical BO approaches that tackle with this scenario.

Seminar by Eduardo César Garrido Merchán at ITEFI on October 9, 2023.


Professor Eduardo C. Garrido Merchán studied at the Universidad Pontificia de Comillas, where he graduated with extraordinary prize in Computer Engineering. He holds a Master’s degree in Artificial Intelligence from the Universidad Politécnica de Madrid. D. Cum Laude in Bayesian Optimization by the Machine Learning group of the Universidad Autónoma de Madrid. Certified by ANECA as Assistant Professor Doctor. His current research interests are, at the methodological level, Bayesian Optimization, Deep Reinforcement Learning, information theory and Generative Artificial Intelligence. At the applied level, Deep Reinforcement Learning and Bayesian Optimization applied to portfolio management and generative artificial intelligence applied to the creation of literary texts and suicide prevention. At the humanistic level, philosophy of mind, philosophy of AI and technology. In 2021 he starts teaching at the Faculty of Economics and Business Administration of the Universidad Pontificia Comillas, where he teaches courses in Machine Learning, Statistics and Quantitative Models.



[Video not available]