Geometric Deep Learning for Inverse Graphics

Student: Kozlukov Sergei

Educational Programme: Statistical Learning Theory (Master)

Final Grade: 7

Year of Graduation: 2020

This thesis deals with geometric methods in deep learning. The main contribution is to refine the theoretical grounds of hyperbolic neural networks. On the experimental side, the thesis proposes two proof-of-concept models that attempt to benefit from hyperbolic representations in problems of point cloud and image classification, discusses the failure of these models, and possible ways to fix these failures. As this work does not (yet) achieve the desired goal of constructing principled symmetry and curvature-aware neural layers applicable in computer vision tasks, the author also considers this thesis a map for the future research.

Full text (added May 27, 2020)

Student Theses at HSE must be completed in accordance with the University Rules and regulations specified by each educational programme.

Summaries of all theses must be published and made freely available on the HSE website.

The full text of a thesis can be published in open access on the HSE website only if the authoring student (copyright holder) agrees, or, if the thesis was written by a team of students, if all the co-authors (copyright holders) agree. After a thesis is published on the HSE website, it obtains the status of an online publication.

Student theses are objects of copyright and their use is subject to limitations in accordance with the Russian Federation’s law on intellectual property.

In the event that a thesis is quoted or otherwise used, reference to the author’s name and the source of quotation is required.

Search all student theses