- 79 views
- 125 downloads
Privacy preserving machine learning
-
- Author(s) / Creator(s)
-
A wide range of application sectors is progressively using machine learning (ML). A successful ML model often requires a huge amount of training data and powerful computational resources. Due to thepotential risks of highly sensitive information being leaked, the need for and use of such enormous volumes of data raise serious privacy concerns. In addition, the changing regulatory environments that increasingly restrict access to and use of privacy-sensitive data present significant obstacles to fully utilizing the power of ML for data-driven applications. There are several techniques for achieving privacy in ML. Homomorphic Encryption (HE) is a public key cryptographic scheme. HE can perform inference on encrypted data, so the model owner never sees the client’s private data and, therefore, cannot leak it. HE is computationally expensive and restricted to certain kinds of calculations. Federated Learning (FL) is a collaborative machine learning method with decentralized data and multiple client devices. During the FL process, each client trains a model on their data set and then sends a model to the server, where a model is aggregated to one global model and then again distributed over clients. Split Learning (SL) is a distributed and private deep learning technique that is used to train neural networks over multiple data sources while mitigating the need to share raw labeled data. This research will provide insight into the trade-off between performance and security for HE, SL, and FL among various ML and DL algorithms.
-
- Date created
- 2022
-
- Subjects / Keywords
-
- Type of Item
- Research Material