Advanced Computer Science (Artificial Intelligence) MSc postgraduate personal statement example

Personal statement example

The wood pigeon nearly sank my final-year project. I had placed a Raspberry Pi with a cheap USB microphone in my parents' garden to record birdsong for a classifier, and for three weeks one pigeon sat on the fence beside it every morning. Its call drowned out everything else, so my 'robin' examples were mostly pigeon with a robin somewhere behind it. Working out how to handle that taught me more about machine learning than any lecture, and it is why I want to study artificial intelligence at Master's level.

My degree is in electronic engineering, so I came to AI through signals rather than software. Modules in digital signal processing gave me a solid grasp of Fourier transforms and filtering, and I took an optional machine learning module in my third year because I wanted to see what happened after the filtering stopped. For my project I built a pipeline that converted short recordings into mel spectrograms and fed them to a small convolutional network. The aim was to identify six common garden species on the device itself, without sending audio to a server.

The pigeon problem was really one of class imbalance and label noise. I relabelled about four hundred clips by hand, separated overlapping calls where I could, and used time-shifting and added background noise to enlarge the rarer classes. I also tried starting from a model pretrained on general audio, which improved accuracy on the smaller classes more than any change I made to my own architecture. Reporting overall accuracy alone hid how badly the model did on wrens, so I learned to look at per-class precision and recall and a confusion matrix. The final system ran on the Pi in under a second per clip. It was modest work, but it was mine from microphone to evaluation, and my supervisor encouraged me to keep developing it.

Since graduating I have worked at a garden centre, on the tills and increasingly on stock. Every Monday I reconcile deliveries against the order system. That has shown me how messy real data is: plants entered under three names, seasonal lines that appear for six weeks and vanish. I built a spreadsheet that flags likely duplicate product codes, which the manager now uses before the spring rush. It is not AI, but it made me think about how any model depends on the record-keeping beneath it.

To prepare for further study I have been working through Russell and Norvig's Artificial Intelligence: A Modern Approach, mainly the chapters on search and probabilistic reasoning, which my engineering course barely touched. I am also reading parts of Goodfellow, Bengio and Courville's Deep Learning to firm up the mathematics behind methods I used somewhat on trust, particularly regularisation and optimisation. I have rewritten my training loop in PyTorch without the high-level wrappers I relied on before, so I understand what each step does.

Outside work I ring bells at my local church. Method ringing is essentially working through permutations of the bells in a fixed pattern, and learning to keep my place in a method has made me patient with long, rule-based sequences. On Saturday mornings I am one of the timekeepers at our parkrun, a small job that matters because a missed button press affects hundreds of results.

A Master's would give me the breadth I currently lack: a firmer theoretical base, exposure to areas such as reasoning and planning alongside learning, and the chance to complete a larger research project under supervision. I am especially interested in machine learning that works with limited, noisy data on small devices, but I want to test that interest against the wider field before narrowing it. I bring practical engineering habits, care with data and a record of finishing what I start.