Work

Employing deep learning to capture and characterise natural behaviours in Bumblebees — Built a deep learning pipeline to discern response to pesticides using video input data. Used a combination of object recognition networks for animal pose estimation and unsupervised learners to classify behaviours.

Image segmentation of mouse intestine slides — Worked alongside a biological imaging startup in SF to obtain a histopathology dataset applied an network architecture for segmenting foreground from background. Achieved pixel accuracy of 98%.

Crohn’s disease diagnosis with MLP classifier — The dataset was comprised of metabolomic biomarkers in the form of GC-MS peaks across elution times from 4 sample types: blood, urine, faecal, and breath. Baseline classifiers were implemented to identify the optimal sample type. To further diagnostic accuracy, an MLP classifier was developed in Keras.