Machine Learning applied to threat detection and galaxy formation

We present findings from several projects across the industrial and academic domains, with a common thread of using machine learning techniques. This work sees a collaboration effort between the security, medical and nuclear imaging company Kromek Group PLC, and the Institute of Computational Cosmology at Durham University. Within the three industrial projects, we present novel threat detection techniques in security and medical imaging. My research and development work on these projects has been utilised and implemented into products in specific fields. The academic project has a global aim concerning predictions for the recently launched Euclid satellite. Kromek related - redacted The academic phase of this thesis sees the prediction of Hα number counts and clustering bias for calibration of the Euclid survey using updated observations. We compare two approaches to the generation of number counts and luminosity functions using the GALFORM semi-analytical galaxy formation model, the full lightcone method and the less computation- ally intensive interpolative method. We find significant improvements in computational speed using the interpolative method over the lightcone method, without sacrificing accuracy. We then take our findings to generate 3000 GALFORM models to use to train a machine learning algorithm (again an artificial neural network) to create an emulator. We then use this emulator in an MCMC parameter search of an eleven-dimensional parameter space, to find a best-fitting model to a calibration dataset that includes local data, and, for the first time, higher redshift data, namely to the number counts of Hα emitters relevant to the Euclid mission.