Novel Approaches for Performance Optimization within Deep Machine Learning

This project would investigate the use of recent advances in a number of areas of deep learning (machine learning) for use in computer vision object recognition systems applied to a specific application or range of tasks and how these can be optimized in terms of training and validation time performance. This will leverage current state of the art techniques from the literature and recent developments within the Durham University team on this and related topics. Validation will be performed against existing benchmark datasets and/or a range of test environment data. Applications could be in automotive visual sensing, robotics, medical imaging, environment reconstruction or generalized "big data" tasks. Final project objectives will be via mutual agreement between the student and the supervision team.