Similarity Learning on Multimodal Data

Dissertation topic at KTIML MFF UK

Similarity learning is a process of learning similarity function that measures how similar or related two objects are. It has various applications in ranking, recommendation systems, visual identity tracking, face verification, speaker verification and product mapping. Product mapping is the process of matching identical products from different e-shops, where each product can be characterised by some of the characteristics as a name, a price, some images, a description, or a specification. The problem is to find matching pairs of products among them.

This project focuses on automatic searching for similarities in multimodal data, such as structured and unstructured texts and different images with applications to product mapping problem. We focus on developing machine learning-based and deep learning-based systems to search for the best metrics that can measure the similarity of objects described by such kind of multimodal data. In practise, these systems can be used for price comparison of a product among several e-shops for further analysis of the e-shops offers.

NAKI: Sources of Krkonoše

Project at UFAL MFF UK

This project is one part of the Program of Applied Research and Development of National and Cultural Identity (NAKI II), provided by the Ministry of Culture. The main intention of the NAKI project is to make information sources about the history and cultural memory of Krkonoše publicly available in one virtual place regardless of the locations of institutions where the original documents are stored. It will be achieved by making a web interface and database for evidence, processing, and presentation of the history of Krkonoše. The output of the NAKI project is a web interface that will enable the creation of metadata descriptions of collected data about the history of Krkonoše and a database for the inserting of documents from the first half of the 20th century and their editing via the web interface.

Multilingual Transfer for Question Answering

Master Thesis at UFAL MFF UK

Question Answering is a computer science discipline in the field of Natural Language Processing and Information Retrieval. Its goal is to build a system that can automatically find the answer to a certain question in the text with an understanding of the meaning of the text. Reading comprehension used for Question Answering is a well-studied task, with huge training datasets in English. This project focuses on building reading comprehension systems for Czech, without requiring any manually annotated Czech training data using text translation and cross-lingual transfer models.

Generator of Nanotubes

Bachelor's Thesis at KAM MFF UK

The clay mineral called halloysite with crystalline structure is widely used in the world of physics due to its effective properties. The main goal of this thesis is to create a program that creates various tubular nanostructures of this material according to input crystal cell and other parameters. After the creation, the whole halloysite structure can be used for other physical research.

Nature Inspired Algorithms (Přírodou inspirované algoritmy) [NAIL119]

Summer semester (Thursday,15:40), 2/2, Z+Zk, 5 credits

In this subject, basic algorithms inspired by nature as Evolutionary Algorithms and Neural Networks and their applications for solving optimization problems and machine learning are presented. During the practicals, some of these algorithms are implemented and used to solve interesting problems in given areas.

The materials for the practicals are available on Github.

The lecture is taught by Martin Pilát and the materials are available on this website.

Theses and Student Projects

Individual Software Project (Ročníkový projekt) [NPRG045]
Both semesters, 0/1, Z, 4 credits

Bachelor's Thesis [NZS031]
Both semesters, 0/4, Z, 6 credits

It is possible to arrange to work on an Individual Software Project or Bachelor's Thesis under my supervision. I am concerned about topics from Artificial Intelligence and Nature Language Processing. If you are interested in any of these topics, you can contact me by email and we can come up with some interesting topics for your project, or I can refer you to other relevant possible supervisors.

Supervised Theses and Projects

2020/2021

• David Nápravník: Software Solution for Digital Archives [Bachelor's Thesis]
• Tomáš Zasadil: Automatic Bird Species Audio Detection [Bachelor's Thesis]