Harnessing the Power of Distributed Computing: Advancements in Scientific Applications, Homomorphic Encryption, and Federated Learning Security

Loading...
Thumbnail Image

Authors

Tawose, Olamide Timothy

Issue Date

2023

Type

Dissertation

Language

Keywords

big data , distributed computing , federated learning , scientific computing , security and privacy

Research Projects

Organizational Units

Journal Issue

Alternative Title

Abstract

Data explosion poses lot of challenges to the state-of-the art systems, applications, and methodologies. It has been reported that 181 zettabytes of data are expected to be generated in 2025 which is over 150\% increase compared to the data that is expected to be generated in 2023. However, while system manufacturers are consistently developing devices with larger storage spaces and providing alternative storage capacities in the cloud at affordable rates, another key challenge experienced is how to effectively process the fraction of large scale of stored data in time-critical conventional systems. One transformative paradigm revolutionizing the processing and management of these large data is distributed computing whose application requires deep understanding. This dissertation focuses on exploring the potential impact of applying efficient distributed computing concepts to long existing challenges or issues in (i) a widely data-intensive scientific application (ii) applying homomorphic encryption to data intensive workloads found in outsourced databases and (iii) security of tokenized incentive mechanism for Federated learning (FL) systems.The first part of the dissertation tackles the Microelectrode arrays (MEAs) parameterization problem from an orthogonal viewpoint enlightened by algebraic topology, which allows us to algebraically parametrize MEAs whose structure and intrinsic parallelism are hard to identify otherwise. We implement a new paradigm, namely Parma, to demonstrate the effectiveness of the proposed approach and report how it outperforms the state-of-the-practice in time, scalability, and memory usage.The second part discusses our work on introducing the concept of parallel caching of secure aggregation to mitigate the performance overhead incurred by the HE module in outsourced databases. The key idea of this optimization approach is caching selected radix-ciphertexts in parallel without violating existing security guarantees of the primitive/base HE scheme. A new radix HE algorithm was designed and applied to both batch and incremental HE schemes, and experiments carried out on six workloads show that the proposed caching boost state-of-the-art HE schemes by high orders of magnitudes.In the third part, I will discuss our work on leveraging the security benefit of blockchains to enhance or protect the fairness and reliability of tokenized incentive mechanism for FL systems. We designed a blockchain-based auditing protocol to mitigate Gaussian attacks and carried out experiments with multiple FL aggregation algorithms, popular data sets and a variety of scales to validate its effectiveness.

Description

Citation

Publisher

License

Journal

Volume

Issue

PubMed ID

DOI

ISSN

EISSN