This meta network of Variational Autoencoders can be used to impute and estimate the conditional distribution of missing data as well as be integrated into classification and regression neural networks. Implemented with Pytorch.

The variation in human sex physiology is far to complicated to be characterized with a binary variable that is fixed at birth. This work studies deep learning methods of constructing a variable that is more predictive of sex associated medical phenomena. Implemented with Python.

Designed and trained a generative model that produces images of skin lesions on individuals with dark skin tones, who are under-represented in data. Implemented with Python.

Estimated policy function for Partially Observable Markov Decision Process for managing rapid decompensation and (septic) shock with Neural Network using MIMIC-III data. Implement with Julia.

Developed deep learning method for separation, and featurization of autofluorescent signal in megapixel resolution, multiplexed immunofluorescence images.

Created synthetic images of tissue samples with and without autofluorescence to validated subtraction algorithms using conditional generative adversarial networks.

Trained and evaluated models to segment and classify multiple types of tissue in patients with high grade gliomas using novel, multimodal data at the time of the research. Implemented with Python.

Developed an interactive visualization of social mobility in the United States using D3.js. Uses are able to upload, merge, and download data as well as fit linear models. Implemented with Javascript. Website.