Friday, July 29 2022
12:00pm - 2:00pm
Masters Presentation
Deep Forest: An Alternative to Deep Neural Networks with Applications in Genetics and Proteomics

Abstract
Machine Learning is paving a new way forward in areas of research including genetics, cancer subtyping and prediction, and proteomics. Models such as Support Vector Machines, XGBoost and Deep Neural Networks (DNNs) are replacing wet-lab methods and proving useful in new situations. But they can be prone to limitations such as (1) sensitivity to parameter tuning, (2) overfitting, (3) low precision, and (4) high type 1 error. This is especially true when applied to the high dimensional data that is often utilized in these fields. This paper examines a deep learning framework called Deep Forest (DF) implemented in the gcForest algorithm in Python that has demonstrated competitive performance in many situations compared to traditional deep learning methods. It applies gcForest alongside XGBoost on a real-world dataset related to chronic obstructive pulmonary disease (COPD). Finally, it offers a brief literature review of five seminal papers that have come out in the last few years that highlight and extend the gcForest algorithmís utility.
Speaker:Gregory Matesi
Affiliation:
Location:Zoom


Download as iCalendar

Done