Machine Learning for Mobile Healthcare

As AI becomes increasingly ubiquitous across industries, there is a growing demand for ML models to be deployed on edge devices, driven by the democratization of AI. However, the decision-making processes of AI systems often exhibit biases, prompting a renewed focus on fairness, particularly in industries prioritizing equitable outcomes such as security surveillance, face recognition, and medical applications like dermatology. This research addresses the need for fairness in mobile healthcare, specifically in dermatology, by developing an Android application for skin disease detection and mobile dermatology assistance in remote areas. While existing AI systems boast high overall accuracies, they often neglect fairness considerations, resulting in subpar performance, especially on datasets representing diverse skin tones. Despite the importance of fairness, most neural network architectures prioritize other metrics, disregarding the need for models to run efficiently on edge devices. To bridge this gap, there is a call for smaller networks optimized for hardware constraints, without compromising fairness. This study explores the paper ”The Larger The Fairer? Small Neural Networks Can Achieve”, presented at the Design Automation Conference – 2022. Which introduces an automatic neural architecture search (NAS) methodology called as Fairness and Hardwareaware Neural architecture search (FaHaNa) for network selection. FaHaNa employs a freezing method to accelerate optimization while preserving fairness, effectively minimizing network size and latency for edge devices. The thesis discusses about the successful application of the FaHaNa framework on Android devices illustrates its potential to democratize healthcare diagnostics across diverse demographic and geographic landscapes, making advanced healthcare solutions more accessible and reducing disparities in medical care availability. This work not only showcases the feasibility of achieving fairness in mobile healthcare applications but also sets a solid foundation for future innovations in the domain of equitable, AI-enabled healthcare solutions.

Author

• Nirmala, Chiranjivan Krishnakumar

• Machine Learning
• Fairness
• Android
• Edge Devices
• Neural Architecture Search
• Artificial Intelligence

Identifier

hdl:1920/14259

OAI identifier oai:identifier

oai:MARS:1920/14259