School of Computer Science and Statistics
Deep learning, in particular Neural Networks (NNs), has achieved state-of-the-art results in many applications in the last decade. However, the way these models work and operate is still not fully understood, and in many ways they are approached as black-box when deployed in practice.Unfortunately, this raises several concerns about their suitability to deal with sensitive … Read more
With the great capability of learning high-level representational features of unstructured data automatically, deep learning is acknowledged as a powerful machine learning algorithm for radio signal processing. For example, convolutional neural networks (CNNs) are leveraged to develop automatic modulation classification frameworks in [1]-[3]. CNN and its combination with Vision Transformers are also used to design … Read more
Federated learning is a paradigm breakthrough shift in AI for data privacy preservation. Unlike conventional artificial intelligence (AI) techniques, federated learning does not require the users to share the data, but only the AI model parameters. As such, federated learning has found notable successes in numerous applications (e.g., Google keyboard, localization, security, and data sharing) … Read more
[MSc Level] This project will involve creating a real-time enfacment system from the state of the art deep-fake networks, and conducting perceptual experiments to identify if participants have ownership of their new face, and if properties of the source-actor (e.g., age, gender, etc.) can be perceived, even though the generated video is a photorealistic depiction … Read more
[MSc level] Virtual assistants are becoming commonplace but the ability for them to gesture naturally and appropriately is still a huge research challenge (see image which shows a typical assistant displayed from the neck up, without gesturing arms and hands). In this project, you will develop a method to automatically learn structure from speech sequences … Read more
[MSc Level] We will collaborate with award-winning animation studio Cartoon Saloon to investigate animated character body part recognition. This project aims to search through cartoon scenes in order to identify character body parts. This task is more difficult than typical human pose-estimation as cartoon characters are often drawn to defy the laws of physics. Knowledge … Read more
The sudden rise of adversarial examples (i.e., input points intentionally crafted so to trick a model into misprediction) has shown that even state-of-the-art deep learning models can be extremely vulnerable to intelligent attacks. Unfortunately, the fragility of such models makes their deployment in safety-critical real-world applications (e.g., self-driving cars or eHealth) difficult to justify, hence … Read more