Amirata Ghorbani, PhD

PhD Advisor: James Y Zou Thesis: Model Interpretation and Data Valuation for Machine Learning - Published 20 papers in venues like Nature, NeurIPS, ICML, ICLR, AISTATS, AAAI. - Completed courses/projects in machine learning, deep learning, natural language processing, optimization, statistics, etc. - Conducted several research projects on Machine Learning Interpretability, examined the drawbacks of existing methods, and created several novel interpretability algorithms for deep learning models. - Created the first-ever algorithm for equitable data valuation in machine learning, show-cased its applications in domain adaptation, data cleaning and preprocessing, and data interpretability. - Designed EchoNet, the first deep learning model capable of performing cardiologist-level detection, measurement, and diagnosis from Echocardiogram videos. Released the largest video medical dataset of more than 10k echocardiogram videos.

-Designed and patented a new algorithm for large batch active learning in deep convolutional neural networks. -Designed a new weak self-supervised algorithm for classification of high-resolution histopathology slides.

-Designed, published, and patented DermGan, a deep-learning-based algorithm for generating diverse synthetic clinical skin images in order to improve the diversity of clinical dermatology datasets with the goal of improving the presence of dark-skinned lesions.

-Designed, implemented, and published the first automatic concept-based interpretability algorithm for Deep Convolutional Neural Networks.

-Ranked 2nd, Class of 2012-2016