MPS Powered by Eureka

Submission Tilte

Homomorphic Data Analysis and Machine Learning

Submission Abstract:

Nowadays, big data systems require processing, representation, and analysis of large amount of digital information. To do that, data processing and mining models, like deep learning ones, rely on complex models that can recognize important patterns inside raw data. The construction of such models involves complex tasks that require substantial computational resources. Hence, high performance and distributed computational platforms stand as necessary solutions to create such models as well as to host pre-trained solutions based on machine learning algorithms. However, such solutions in general have some drawbacks related to data privacy, especially in the case of cloud platforms. For instance, in deep learning as a service in cloud environments, the user can upload the input data to the cloud which in turn evaluates a pre-trained neural network algorithm on the input data and sends the results back to the user. Along this process, attackers have several opportunities to intercept the computation to access (or compromise) sensitive data. This issue is more critical in federated learning where a central model is decentrally trained through a network of servers, each one hosted in some device. Hence, the research in the subject of privacy-preserving data processing and mining algorithms become a fundamental area to allow the application of high performance and cloud solutions. In this context, a reliable solution would be to compute on encrypted data without decrypting it, a solution obtained with (fully) homomorphic encryption schemes. The main goal of this special issue is to explore homomorphic encryption techniques for data processing and data analysis in pattern recognition tasks. We encourage original submissions in machine/statistical learning methods working in homomorphic data representation for model training and/or extracting meaningful information from high-dimensional data spaces with the guarantee of preserving data secrecy. Review articles that describe the current state of the art in Homomorphic data Processing are welcome as well. Papers are published upon acceptance, regardless of the Special Issue publication date.