Research Projects

Federated Learning (FL) has emerged as a promising approach for collaborative machine learning, allowing multiple clients to jointly train a global model without sharing their local data. However, current FL methods often lack measures to prevent discriminative or unfair predictions. An FL setup is more vulnerable to fairness issues due to the distributed training regime and independent clients who could be heterogeneous, strategic over even malicious. Given the challenges of non-IID data, free riders, and malicious clients, this research aims to design, develop, and evaluate a federated learning framework that ensures fairness. Specifically, we aim to, • Analyze and Benchmark: For different datasets and state-of-the-art approaches provide fairness and performance guarantees under the presence of o Heterogeneous clients or non-IID data o Strategic clients who are trying to free-ride o Malicious clients who are trying to introduce bias • Develop a Fair Aggregation Algorithm: Design a novel aggregation algorithm that can effectively combine updates from heterogeneous clients, considering the diversity and distribution of their data. The algorithm should compensate for the Non-IID nature of the data to ensure a fair and balanced global model. • Incorporate Incentive Mechanisms: Implement game-theoretic incentive mechanisms that encourage fair participation. These mechanisms should identify and reduce the influence of free riders, ensuring that contributions reflect in the shared model's performance and benefits. • Design Robust Detection Schemes: Create detection schemes capable of identifying and mitigating the impact of malicious clients. These schemes should be able to handle sophisticated data and model poisoning attacks without compromising the learning process. • Evaluate Framework Performance: Conduct comprehensive evaluations to assess the framework’s effectiveness in a federated learning environment. This includes testing the framework against various degrees of data heterogeneity, different strategies of free riders, and a range of security threats. The successful completion of this project will contribute to developing a fair and secure FL framework that can be applied in various domains, such as healthcare, finance, and personalized recommendations, where data privacy and fair decision-making are of utmost importance. The proposed solutions will enhance the practicality and trustworthiness of FL, promoting its wider adoption in real-world settings. For example in healthcare, it would promote Personalized medicine, by enabling training of models on decentralized patient data for personalized diagnosis, treatment recommendations, and drug discovery, while ensuring fairness and preventing biases based on demographics or data imbalances. It also enable collaborative analysis of medical records and/or images from various hospitals to predict disease outbreaks and develop preventative measures without compromising patient data.