PhD position: Deep Learning for Image and Video Understanding

Advisors: prof. Oswald Lanz and prof. Sergio Escalera

Further information: please contact advisors

Application instructions:
https://www.unibz.it/en/faculties/computer-science/phd-computer-science/

Project Description:
The project will explore novel techniques for semi-automated design and effective training of deep architectures for image and video understanding. Of particular interest are scenarios where lots of raw data is available but either annotation is scarse and/or distribution of data is highly unbalanced. Such situation is often encountered in anomaly detection, recognition and forecasting, preventing supervised learning at scale. Transfer learning, data augmentation and a task-specific architecture design can alleviate but this typically requires the inventive contribution of an expert. Investigations in this project will contribute towards automating the design and training process by leveraging self-supervised pre-training, neural architecture search, multi-modal learning. Furthermore, domain knowledge is often available but it is non-trivial to represent and integrate with deep learning. Relevant research contributions might include knowledge injection and distillation, and neuro-symbolic integration for vision. The application context for this project is not sharply defined, but can build upon previous research of the group (see http://srl4v.github.io) and the student can consider use-cases from industry (see https://covisionlab.com).

Required skills:
The eligible candidate has undertaken computer vision and deep learning studies at university courses with proficiency. The MSc thesis is in the field of deep learning and computer vision, ideally already providing the background on one aspect related to the project and the research proposal by the applicant. Solid programming skills and experience with deep learning frameworks such as pytorch are requested.