20430032 - MATHEMATICAL METHODS FOR MACHINE LEARNING

Illustrate some of the mathematical methods that are at the basis of Machine Learning, and in particular linear algebra, convolution, minimization, and the structure of Neural Networks.
teacher profile | teaching materials

Programme

Review of the fundamentals of linear algebra, special matrices
Matrix factorization and decomposition; eigenspaces
Large matrix management and data reconstruction
Optimization and Lagrange multipliers
Minimization and gradient method
Structure of Neural Networks (NN)
Fundamental building blocks of NNs: affine layer; activation layer; cost layer (quadratic loss and cross entropy)
Backpropagation
Using NNs: Data split; Minibatch; Training & Validation
Classification and regression
Image analysis and convolution
Matlab: NN assembly; performance analysis; sample problem selection
Professional NNs

Core Documentation

Gilbert Strang, Linear algebra and learning from data, WELLESLEY- Cambridge Press.

Michael Nielsen, Neural Networks and Deep Learning (web book).
http://neuralnetworksanddeeplearning.com/index.html

The material will be indicated by the instructor during the lectures

Attendance

Classroom lectures and exercise sessions

Type of evaluation

1) Solve exercises from Strang's book; the exercises should be submitted via email 2) Prepare a report on a topic of your choice, agreed upon with the instructors 3) Arrange a presentation appointment with the instructors 4) Submit the exercises and the report at least one week before the presentation 5) Discuss the report in a satisfactory manner In evaluating the project, the grade will take into account several factors, such as: the level and quality of the knowledge of the topics presented; the quality of the report submitted; the ability to critically analyse the topic; the ability to apply theories and concepts to the project.