Senior machine learning research scientist at Imprint Labs, building generative and unsupervised models of the adaptive immune system.
I work on how the adaptive immune system recognizes pathogens, and on reading that from sequencing data. The data are indirect and noisy, so much of the work is handling them well: parsing them carefully, modeling how they were measured, and being clear about what they can and cannot tell us. Flexible models, neural networks included, have to earn their place against simpler ones.
One example is soNNia, a public package for modeling selection in T and B cell receptor repertoires. I also co-wrote a review of machine learning for adaptive immune receptor analysis.
How the principle of maximum entropy connects unsupervised learning and density-ratio estimation, and its surprising link to logistic regression.
Read →Calibrating model predictions to data by matching quantiles with boosted decision trees, extended to the multidimensional case.
Read →One linear model of neutral allele frequencies reads transmission networks from SARS-CoV-2 and ancient migrations from archaeogenetic data.
Read →