Research Project Title:
Confidence Estimation for Molecular Property Prediction
abstract:In Professor Barzilay's lab, I will be working to develop new features for ChemProp, a molecular property prediction tool which uses machine learning to offer improved performance when compared to traditional methods, like circular fingerprints. One such addition is the measurement of confidence intervals, which can assist in estimating a model's accuracy for any particular input. While there exists no consensus as to how to best calculate these values, I hope to test the efficacy of a number of proposed methods with ChemProp and so, select those that are most competitive. I will also work to improve the model's ability to transfer to new domains.
"I am very excited to further develop my research skills by participating in this SuperUROP. I am especially interested in the work done in Professor Barzilay's lab because I believe it to be both high impact and technically challenging. Beyond expanding my knowledge of deep learning, an area I've focused on, this research will also help boost the efficiency of workers in the pharmaceutical industry."