Neural Networks allow creation of programs that are learned rather than written. This means that instead of implementing a concrete algorithm, the program applies patterns that are automatically found in the data. For example, a translation system such as Google Translate can be created by applying a deep neural network to a large corpus of translated documents. The aim of this course is to present the basics of how neural networks work, both from a more theoretical and practical point of view. Neural Networks can be applied to a wide spectrum of tasks, including image and video processing, creating game playing agents, financial data analysis, or even using a computer as a writer, painter or composer. We will try to address many such issues with the emphasis on natural language processing.
The lecture can be treated as a continuation of the following courses: Machine Learning, and Natural Language Processing.
The first part of this course is shared with the Neural Networks and Deep Learning. However, students who have finished NN-DL can enroll in this course, and obtain 50% of ECTS points (there will be special rules for them, allowing them to concentrate only on the new material).
The lecture will be accompanied by computer exercises, including small exercises illustrating key topics, slightly more demanding tasks giving better understanding of neural networks and a larger project that will give you the opportunity to tackle a real world machine learning problem.
We will introduce the PyTorch deep learning framework, and PyTorch-based popular NLP libraries (Flair, AllenNLP)
This course will have two lecturers: Rafał Nowak, for the first part and Paweł Rychlikowski for the second part.
