Introduction
The ML Systems & Optimisation research area focuses on the design and optimisation of machine-learning models and their execution across different computing platforms.
Research includes efficient model architectures, model compression, automated design-space exploration, distributed and parallel training, and optimisation of inference for embedded, FPGA, cloud, and other resource-constrained environments. The work considers model accuracy together with computational cost, memory use, latency, throughput, and deployment constraints.
A recurring objective is to treat the machine-learning model and its execution environment as a coupled design problem, allowing algorithmic and system-level choices to be explored together.
Research Directions
Efficient model design and optimisation
Design and optimisation of machine-learning architectures with emphasis on accuracy, computational cost, memory use, and inference efficiency. This includes model compression, architecture optimisation, and automated design-space exploration.
Parallel and distributed ML systems
Methods for improving the training and execution of machine-learning workloads across parallel, distributed, and cloud computing environments. Research includes model parallelism, workload decomposition, and scalable execution strategies.
Embedded and FPGA deployment
Optimisation and deployment of machine-learning models on embedded and FPGA platforms, with emphasis on latency, resource utilisation, throughput, and implementation constraints. This includes hardware-aware model design and the evaluation of reduced-complexity architectures for resource-constrained inference.
Related projects
- MergeOver
Selected Artefacts
Student Projects
Student work in this area includes model optimisation, data efficiency, automated design exploration, model parallelism, FPGA synthesis, edge deployment, and high-performance simulation. See the Student Projects page for current opportunities and completed research.