Goals of the Seminar:
- provide an overview of current mathematical methods used in current machine learning,
- provide knowledge of current computational methods used, such as convolutional networks and deep learning,
- gain practical experience in applying machine learning to standard classification problems,
- program in python using keras and/or pytorch machine learning libraries,
- using GPU-processing for deep network training,
- application to problems from speech and audio signals, and to self-chosen problems.
Structure of the course:
First half (weeks 1 to 7) of the course:
We will provide short lecture segments as an introduction to advanced methods from machine learning
relevant to this course. In particular, this will include convolutional networks and several deep network architectures.
We will also provide an introduction to the relevant programming libraries in python that are used, such as keras and pytorch.
Students will work in a self-paced way on a set of python notebooks that introduce these concepts and that include simple implementation steps.
Second half (weeks 8 to 14) of the course:
Students will work individually or in groups on a self-chosen problem in the setting of a mini-project.
The extent of a mini-project will be limited in size and it will follow the implementation practice learned during the first half of the course.
Project progress, necessary technical steps and possible problems encountered will be addressed at regular meetings.
Examples of projects students worked on during previous courses:
- Music genre classification
- Emotion recognition from speech
- Music melody generation
- Natural language processing for tweets
- introductory course to machine learning, signal processing etc.,
- basic knowledge of python programming,
- (ideally) knowledge of jupyter notebooks,
- (ideally) knowledge of linux.