Uncertainty and Capsule Networks for Computer Vision

<p>Deep learning is a particular kind of machine learning which is powerful andflexible as a consequence of its ability to represent the world as a nested hierarchyof concepts (LeCun et al. 2015, Goodfellow et al. 2016). Due to the ever increasingpower of parallel computing graphics processing units, larger labelled datasets andimproved training techniques, great leaps in the performance of various machinelearning tasks have been achieved using deep learning (LeCun et al. 2015). At thetime of this writing, deep learning is the dominant machine learning approach formuch ongoing work in fields such as: Computer Vision (Krizhevsky et al. 2012, Heet al. 2016), Reinforcement Learning (Mnih et al. 2015, Silver et al. 2016), MedicalImaging (Ronneberger et al. 2015), and Natural Language Processing (Vaswaniet al. 2017, Devlin et al. 2019).However, with the uptake of deep learning models into safety-critical domains,transparency of model predictions is becoming increasingly important for: safety,decision-making, fairness and legislative reasons. Moreover, designing deep learning models that strike a good balance between human interpretability and performance has proven to be a challenging task (Caruana et al. 2015, Montavon,Lapuschkin, Binder, Samek & Müller 2017, Kendall & Gal 2017, Rudin 2019,Samek et al. 2019). With that said, in this thesis we advocate for an alternativeview of interpretability based on estimating the uncertainty in a model’s predictions, which serves as a proxy for model transparency. In our investigations,we formalise the desiderata of model transparency as: trust, information andgeneralisation, and take steps towards the development of deep learning modelswhich have the potential to satisfy them. Concretely, we leverage the language ofuncertainty to improve the performance and transparency of deep learning models in computer vision tasks, providing probabilistic techniques to enhance moreinterpretable models by design such as capsule networks.</p>