Course Description
The information present in Web and multimodal data allow the creation of systems capable of answering diverse and complex user information needs. Use cases include, but are not limited to, the medical domain and in the social media domain. The objective of this course is to allow students to understand all aspects related the representation, extraction and access to Web and multimodal information. Students will learn the main techniques to create multimodal search spaces, indexing through similarity-based hashing and annotation of information to allow advanced search.
The course program is divided into four parts:
- Part 1 - Introduction to metric spaces and approximate NN
- Part 2 - Information extraction
- Part 3 - Graphs
- Part 4 - Learning high-level embeddings
A project will be developed
throughout the semester, allowing the student to consolidate the
studied techniques into
a single Web/multimodal search system targeting a specific
information domain, e.g., medical, video surveillance and social
media.
Objectives
- Understand the concept of error in information representation.
- Understand Web and multimedia representation models.
- Analyze Web and multimedia data for information extraction.
- Understand Web and multimedia information access paradigms.
Grading
Exam (40%) + Project (60%, divided in 3 steps)