 |
Mauro DELL'AMICO
Professore Ordinario
Dipartimento di Scienze e Metodi dell'Ingegneria
|
Insegnamento: Automated Decision Making
Ingegneria informatica (Offerta formativa 2024)
Obiettivi formativi
The goal of the course is to examine recent topics in the application of machine learning techniques to the solution of computational problems in discrete optimization and vice versa, to analyze the use of discrete methods to enhance machine learning techniques.
Tools from ML, mixed-integer programming, and heuristic search will be studied, analyzed, and applied to a variety of discrete optimization problems such as, the traveling salesman problem, vehicle routing, graph coloring, etc.
The course includes laboratory activities using Python, PyTorch (or your preferred neural network platform) and MIP solvers as Gurobi
Prerequisiti
Knowledge of machine learning techniques, programming, and operations research
Programma del corso
The first partof the course will cover background material, resuming machine learning techniques ( deep learning) focusing on practical aspects, major topics in applied mixed-integer programming, including modeling, such as, linear-programming duality, cutting planes, column generation, and branch-and-bound, (1CFU) and will introduce heuristic search techniques (1CFU). We will then make an in-depth study of recent papers where machine learning has been proposed as a solution-technique in discrete optimization and where mixed-integer programming has been adopted to study the performance of deep learning models (4CFU) The presentations will be aimed towards discussions of open research questions.
Metodi didattici
Seminars, homeworks, a major project appling ML to a discrete optimization problem.
Testi di riferimento
Deep Learning, Goodfellow, Bengio, and Courville
In Pursuit of the Traveling Salesman
Model Building in Mathematical Programming, H. Paul Williams
Verifica dell'apprendimento
Homework 20%
Major Project 65%
Final discussion on the projects and course subjct 15%
Risultati attesi
Capacity to integrate discrete optimization and machine learning algorithm to solve complex problems.