Human Inspired Video Recognition using top down and bottom-up features

How does the human brain understand people's activities in a scene much better than existing computer systems? It is proposed that humans combine prior knowledge of people and objects in the scene with sensory information (e.g., colour or motion) to understand complex activities even if they are similar actions (e.g., hugging vs fighting).

The project collaborates with a Psychology PhD student at the Biosciences Institute, Newcastle University, to help answer these questions. You will create more explainable computer models to recognise video activity categories to help refine our understanding of how the brain works for recognising everyday human activities. Later you will use the human-inspired weights and identified critical features to re-train the computer models and test whether they improve accuracy on the videos that are hard to discriminate.

To disseminate the knowledge you gain, we expect you to lead and publish several first-author publications in high-quality, internationally renowned Machine learning (ML) conferences, such as CVPR, ICCV and ECCV. We also expect you to collaborate with Newcastle on Journal publication, e.g., in Cognition.

All applicants should have (or expect to obtain) a first-class degree in a numerate discipline (mathematics, science or engineering) or MSc with Distinction (or 70% average) and a strong interest in pursuing research in this field. Additional experience which is relevant to the area of research is also advantageous. Previous experience in programming is essential. Previous experience in computer vision and deep learning would be valuable.

You will need to meet the minimum entry requirements for our PhD programme.

Non-native English speakers must have IELTS 6.5 or above (or equivalent) with no sub-test of less than 6.