Matt Bowditch

I am a PhD student based at the University of Warwick in the Mathematics for Real-World Systems CDT, supervised by Prof. Ranko Lazic and Dr. Matthias Englert.

My current research interests include topics based around convolutional neural networks and image-diffusion models such as:


Past Projects

Embedding layer backdoors on siamese neural networks

This project was completed as part of my Mathematics of Systems MSc, supervised by Prof. Ranko Lazic and Dr. Matthias Englert. This work was based on a backdoor that can be implemented in facial recognition systems which forces the system to err on attacker-specified people whilst only losing a small amount of accuracy in identifying others (Shamir & Zehavi, 2023). This can be used to anonymise a person or to cause the system to confuse two people as being the same person.

This work extended on results from the aforementioned paper by:

I am continuing this work into my PhD.

Evolving predator-prey dynamics in agent-based models of collective motion

This was a group project completed as part of my Mathematics of Systems MSc, supervised by Prof. Matthew Turner and Prof. Gareth Alexander. We proposed a force-based model of agents to replicate the dynamics of a group of predators attacking a swarm of prey in a bounded space, involving novel approaches to boundary conditions and predator evasion.

We then implemented an original evolutionary adaptation mechanism on the prey and predator behavioural parameters, to minimise or maximise the proportion of prey killed respectively. The parameter optimisation process was carried out using the BIPOP-CM-AES algorithm sequentially. Optimisation resulted in continual oscillation between distinct strategies, without the emergence of a dominant strategy for either species.

The cavity method applied to group-structured networks

This project was completed as part of my MMath degree from the University of Bath. It was supervised by Prof. Tim Rogers. This extended on work regarding the use of the cavity method to approximate individual node risk when an epidemic is spreading through the network (Moore & Rogers, 2020). Such methods give a good approximation when networks are tree-like, however, the accuracy decreases when more structured networks are considered.

This project proposed adaptations to this methodology in order to give more accurate approximations in cases in which networks contain many small fully connected subsets of nodes. An analysis of the method was provided with useful properties being determined, such as the contagion threshold: the maximum amount of transmissibility before there is a risk of a large-scale outbreak in the network.

Education & Employment

2023-present: University of Warwick

PhD in Mathematics of Systems

2022-2023: University of Warwick

MSc in Mathematics of Systems, distinction

2021-2022: Screwfix

Customer data analyst

2017-2021: University of Bath

MMath, first-class

Hallo, Welt

Das hier ist nur ein kleines Beispiel von einem Artikel.