Keio University Syllabus and Timetable

This is an introductory course on modern Artificial Intelligence designed for Keio University. The course is composed of two parts taught in consecutive semesters. The material introduced in part A forms a foundational basis for part B (this course), which develops these ideas further and introduces a selection of more recent results based on guided reading of relevant publications. The two courses taken in sequence form a coherent introduction to neural Artificial Intelligence. The first course focuses more on theory and fundamental concepts, with implementation of basic techniques in Python. The second course (this one) aims to cover more practical engineering topics using modern practices, as well as introducing some of the most influential recent advancements based on a selection of research papers. Part B of the course also introduces some topics in more depth, based on the interests of the instructor. One of those topics is Natural Language Processing (NLP) in the era of Deep Learning, as well as advanced methods in representation learning.

Recent years have brought a revolution in the field of Artificial Intelligence on an unprecedented scale. Advances in hardware, availability of large data sets, as well as innovation in architectural and algorithmic design, enabled successful application of Machine Learning models based on multi-layered Artificial Neural Networks to a variety of problems of practical interest. The subfield of AI focusing on deep (multi-layered) neural architectures and the associated algorithms is collectively known as Deep Learning. The success of Deep Learning has been so great in recent years, that most of modern AI can be summarized as the study of deep neural architectures. Deep Learning models hold state-of-the-art (SOTA) results on virtually all AI tasks, and new discoveries are made almost daily.

In many real world Machine Learning tasks, in particular those with perceptual input, such as vision and speech, the mapping from raw data to the output is often a complicated function with many factors of variation. In past approaches, to achieve decent performance on such tasks, significant effort had to be put to engineer hand-crafted features. Deep Learning algorithms aim to learn feature hierarchies with features at higher levels in the hierarchy formed by the composition of lower level features. This automatic learning of abstraction has been demonstrated to uncover underlying structure in the data (c.f. manifold hypothesis). The first successes came in the fields of vision and speech processing. Recently, NLP, and reinforcement learning became the new frontiers for deep neural networks.

This course aims to introduce students to the field of Artificial Intelligence, focusing on Deep Neural Information Processing Systems. Since this is a rapidly developing field, we will focus on most important trends and core ideas, as it is impossible to cover all recent developments in a single course. We will follow historical trends in AI with a focus on neural networks, and see how the current ideas emerged out of decades of research in the field. We will then discuss current neural architectures and algorithms, and introduce modern perspectives. After completing this course, students should have appreciation and understanding of neural AI systems, and anticipate future developments in research and applications of AI, and Deep Learning in particular. In addition to theory, there will be emphasis on programming skills in Python. We will implement deep neural AI systems, and train them on standard data sets.

In this course, will discuss a selection of more complex neural information processing systems used in state-of-the-art results. We will also develop programming skills by implementing theoretical ideas in code, cover mathematical and algorithmic notions relevant to understanding Deep Learning systems, and engage in guided reading of scientific literature. It is recommended that students complete both courses (A and B) in sequence. However, some students could take this course in isolation, after consulting the instructor during the first lecture. In such cases, students should review the material from part A on their own time, as this course will build on previously introduced concepts.

This item will appear when you log in (Keio ID required).

This item will appear when you log in (Keio ID required).

Reading materials will be provided by the instructor.

Reading materials will be provided by the instructor.

This item will appear when you log in (Keio ID required).