Towards uncertainty-aware and privacy preserving deep learning

Thesis

Deep Learning has revolutionized numerous fields, achieving state-of-the-art performance in areas like computer vision, natural language processing and applied sciences. This progress has led to its integration into increasingly critical applications, with Deep Neural Networks already guiding crucial decisions in areas like autonomous vehicles, financial trading, mortgage assignment, hiring procedures, weather forecast, medical diagnosis, satellite management etc. Due to their intricate struc...

Expand abstract

Authors: Pinto, F

• DOI: 10.5287/ora-6r9erznnj
• Type of award: DPhil
• Awarding institution: University of Oxford
• Language: English
• Keywords: adversarial attacks; multimodal models; uncertainty estimation; large language models; differential privacy; image classification