Studies on biomolecular structure learning and federated privacy protection
Abstract
This thesis mainly focuses on two topics: 1) elucidating the structure and dynamics of biological macromolecules, 2) privacy protection of federated learning. Understanding the structure and dynamics of biological molecules is a central pillar of molecular biology. A series of machine learning tools are developed in simulation or experimental data. In particular, one part of the work involves better utilization of Cryo-EM images obtained experimentally. A robust denoising approach is proposed in Cryo-EM images to help structure reconstruction. Then a two-stage classification scheme is presented that could accurately classify the denoised Cryo-EM images from multiple conformations, which allows inferring the underlying free energy landscape. In addition, a kinetic algorithm based on projection operator framework via deep learning is designed to assign numerous microstates into a handful of metastable states. It is helpful to understand the conformational dynamics of complex biomolecules from the simulation perspective. Another work focuses on federated privacy protection. Federated deep learning aims to preserve users' data privacy by decentralizing data from the central server to end devices. However, adversaries may still infer the private training data from the released model and updated gradients. A privacy-preserving Federated Deep Learning with Private Passport framework is proposed without sacrificing the model performance and computation efficiency.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.