At IMU we are creating an extraordinary platform to map and decode the human immune system. We turn 2ml of blood into 5M data points using spectral flow cytometry, facilitating unprecedented resolved insight into the cell populations and states that govern our biological immune defense. With machine learning we are vastly simplifying these complex data structures and visualizing human cell networks by applying our data science toolbox in new and innovative ways.

My role as a postdoctoral fellow involved three core responsibilities.The first was using supervised machine learning to assist with evaluating data quality:• Multiple supervised machine learning projects to evaluate data quality. Heavily depended on by collaboration of ~100 globally-distributed team members working on $42M experiment. • Delivered innovative models and algorithms to meet the requirements of the collaboration using advanced Python scripting (NumPy… Show more My role as a postdoctoral fellow involved three core responsibilities.The first was using supervised machine learning to assist with evaluating data quality:• Multiple supervised machine learning projects to evaluate data quality. Heavily depended on by collaboration of ~100 globally-distributed team members working on $42M experiment. • Delivered innovative models and algorithms to meet the requirements of the collaboration using advanced Python scripting (NumPy, scikit-learn, keras, PyTorch) to analyse large volumes of data (TBs).• Excellent interpersonal and communication skills as demonstrated by consistently translating stakeholder domain knowledge into defined data science tasks.• Developed auto-encoder-based supervised machine learning pipeline to classify complex and high-dimensional image data to evaluate quality of ~2M data samples in Python.• Saved experts ~2 months work of manually classifying data: >50x speed up and cost reduction.The second was using unsupervised machine learning to detect anomalies and rare objects:• Designed end-to-end data pipeline in Python including preprocessing, PCA and t-SNE.• Unveiled and characterised non-linear relationships hidden in the data using clustering (e.g. K-means, DBSCAN), adding value to the overall data-set.• Led a small collaborative team to use an isolation forest to formally identify anomalies, results presented to international conference.Finally, the third was an autonomous research position to aiming to understand exploding stars:• Designed and executed large-scale astronomical experiments to collect and analyse time-series data.• Utilised APIs in Python to retrieve large time-series data sets pertaining to objects of interest.• Advanced data analysis and statistical inference in Python including mathematical modelling, Monte Carlo simulation and Bayesian modelling.• Communicated key results to varied stakeholders through presentations and academic reports. Show less

I used Python and its various numerical and scientific libraries to clean and analyse complex data sets from the Dark Energy Survey to learn about superluminous supernovae. I designed a novel set of algorithms in Python using dedicated supercomputing facilities to compute the rate of superluminous supernovae across the cosmos. A special highlight of my PhD was being invited as a plenary speaker at the international Dark Energy Survey collaboration meeting in Campinas, Brazil.• Good… Show more I used Python and its various numerical and scientific libraries to clean and analyse complex data sets from the Dark Energy Survey to learn about superluminous supernovae. I designed a novel set of algorithms in Python using dedicated supercomputing facilities to compute the rate of superluminous supernovae across the cosmos. A special highlight of my PhD was being invited as a plenary speaker at the international Dark Energy Survey collaboration meeting in Campinas, Brazil.• Good knowledge of SQL for accessing and combining data from multiple sources as demonstrated by obtaining time-series data on a rare type of "superluminous" supernova.• Led a team of researchers to manually classify a new sample of superluminous supernova. • Performed complex statistical analysis to determine the rate of superluminous supernovae across the cosmos.• Unsupervised machine learning to identify rare (less than 1 in a million) objects.• Invited speaker at an international scientific conference to present results to ~300 stakeholders. Show less