Multiple Information Source Bayesian Optimization - Andrea Ponti

. .

Résumé :
The book provides a comprehensive review of multiple information sources and multi-fidelity Bayesian optimization, specifically focusing on the novel Augmented Gaussian Process methodology. The book is important to clarify the relations and the important differences in using multi-fidelity or multiple information source approaches for solving real-world problems. Choosing the most appropriate strategy, depending on the specific problem features, ensures the success of the final solution. The book also offers an overview of available software tools: in particular it presents two implementations of the Augmented Gaussian Process-based Multiple Information Source Bayesian Optimization, one in Python -- and available as a development branch in BoTorch -- and finally, a comparative analysis against other available multi-fidelity and multiple information sources optimization tools is presented, considering both test problems and real-world applications. The book will be useful to two main audiences: 1. PhD candidates in Computer Science, Artificial Intelligence, Machine Learning, and Optimization 2. Researchers from academia and industry who want to implement effective and efficient procedures for designing experiments and optimizing computationally expensive experiments in domains like engineering design, material science, and biotechnology. ...

Biographie:
Francesco Archetti is Professor Emeritus of Computer Science and full Professor of Computer Science at the Department of Informatics, Systems and Communication (DISCo), University of Milano-Bicocca, Italy. His research activities are focused on Data Analytics, Network Science, Probabilistic Modelling, Predictive Analytics, and Optimal Learning, with application to security, water management, logistics, and cyber-physical systems. He is one of the two authors of the Springer Brief Bayesian Optimization and Data Science (2019). Antonio Candelieri is an Associate Professor for the Department of Economics, Management, and Statistics at the University of Milano-Bicocca, Italy. His research activities are focused on Machine Learning and Bayesian Optimization. He was ranked within the Top 2% Scientists, Stanford University Ranking 2023 and received a Paper Award 2022 Honorable Mention from the Journal of Global Optimization (Springer). Andrea Ponti is a PhD candidate at the Department of Economics, Management, and Statistics, University of Milano-Bicocca, Italy. His research focuses on the optimization of black-box functions using advanced Bayesian methods. From an industrial perspective, he designs and develops versatile machine learning solutions, focusing on foundation models and Large Language Models (LLMs, aka what's behind ChatGPT). Andrea Ponti is a PhD student in Data Science with a master&rsquo...

Sommaire:
The book provides a comprehensive review of multiple information sources and multi-fidelity Bayesian optimization, specifically focusing on the novel Augmented Gaussian Process methodology. The book is important to clarify the relations and the important differences in using multi-fidelity or multiple information source approaches for solving real-world problems. Choosing the most appropriate strategy, depending on the specific problem features, ensures the success of the final solution. The book also offers an overview of available software tools: in particular it presents two implementations of the Augmented Gaussian Process-based Multiple Information Source Bayesian Optimization, one in Python -- and available as a development branch in BoTorch -- and finally, a comparative analysis against other available multi-fidelity and multiple information sources optimization tools is presented, considering both test problems and real-world applications. The book will be useful to two main audiences: 1. PhD candidates in Computer Science, Artificial Intelligence, Machine Learning, and Optimization 2. Researchers from academia and industry who want to implement effective and efficient procedures for designing experiments and optimizing computationally expensive experiments in domains like engineering design, material science, and biotechnology. ...