Neural Architecture Design
Principled approaches to architecture design — attention mechanisms, normalisation schemes, and connectivity patterns that improve accuracy and efficiency.
Calibration & Uncertainty
Theoretical and practical advances in neural network calibration — making confidence scores reliable for high-stakes decision support systems.
Certified Robustness
Provably robust neural networks — training methods and architectural techniques with verifiable guarantees against adversarial perturbations.
Efficient Neural Inference
Hardware-aware model compression — novel quantisation schemes, structured pruning, and dynamic networks that adapt computation to input complexity.
Continual & Few-Shot Learning
Neural systems that learn efficiently from limited data and continue learning without catastrophic forgetting of prior knowledge.
Scientific ML
Physics-informed neural networks and neural operators for scientific computing — bridging the gap between deep learning and traditional simulation.
Research Output
Frontier research with real-world deployment impact.
Research Collaboration
We partner with universities and research labs on joint publications and funded research programmes.
Collaborate