By Jude Senko

As industries race to adopt artificial intelligence (AI) and machine learning (ML), one of the biggest barriers is protecting sensitive data while still enabling powerful model training. 

For Ph.D. candidate and data engineer Joy Nnenna Okolo, the answer lies in Federated Learning in Mobile Edge Computing (MEC) — a privacy-preserving approach that could change how AI operates across sectors.

Okolo’s research addresses a growing problem: traditional AI models require centralising data in one location, creating security vulnerabilities and privacy risks. In sensitive industries like finance, healthcare, and IoT, this model is both risky and, in some cases, non-compliant with data protection laws.

“Centralising data increases the attack surface for cybercriminals,” Okolo explained. “Federated learning allows the model to learn from multiple devices without ever moving the data, which significantly reduces privacy risks.”

Her work focuses on deploying these federated learning models within mobile edge computing environments — systems that process data closer to its source rather than in distant cloud servers. This reduces latency, enabling real-time decision-making while safeguarding user privacy.

In financial services, for example, Okolo’s framework could allow multiple banks to collaboratively improve fraud detection models without exposing their customers’ transaction histories. In healthcare, it could enable hospitals to develop shared diagnostic AI tools without violating patient confidentiality.

A major challenge she addresses is optimising communication efficiency in federated learning systems. Transmitting model updates between edge devices and central servers can be resource-intensive. Okolo’s research explores compression techniques, adaptive update scheduling, and security protocols to make the process both faster and safer.

She also investigates adversarial attack resilience — ensuring that malicious participants cannot poison the learning process with corrupted updates. This is especially important in open or multi-party federated systems where trust cannot be assumed.

Her work builds on her earlier experience integrating behavioural science with cybersecurity strategies, giving her a unique lens on both the technical and human factors involved in secure AI deployment.

Federated learning is still a developing field, but Okolo’s research offers a practical roadmap for real-world adoption. By combining MEC’s low-latency processing with AI’s adaptability, she aims to make privacy-preserving AI not just a theoretical concept but an industry standard.

The applications extend well beyond banking and healthcare. Retailers could use similar systems for demand forecasting across stores without sharing proprietary sales data. Smart cities could optimise traffic flow without tracking individual drivers.

Okolo believes this approach is crucial for public trust. “People are more likely to embrace AI when they know their data never leaves their control,” she said.

Her next steps involve testing the scalability of her models in simulated multi-organisation environments, with the goal of deploying them in live industry trials.

“We don’t have to choose between innovation and privacy,” she added. “With the right architecture, we can have both.”