Ieee federated Machine Learning White Paper
Download 0.56 Mb. View original pdf
|
e-Iraq estra.ar.en
- Navigate this page:
- 3.3. HEALTHCARE
- 3.4. EDUCATION
| 3.2. TELECOMMUNICATIONS Mobile devices equipped with neural network processing units exploit their strong computational power to train NN models using data captured by a wide range of on-device sensors. With such on-device computational power and data, mobile applications have significantly improved their usability and bring convenience to people’s lives. Authorized licensed use limited to University of Malta. Downloaded on December 24,2022 at 11:03:39 UTC from IEEE Xplore. Restrictions apply. Copyright © 2021 IEEE. All rights reserved. 11 IEEE SA However, serious ethical and regulatory concerns about data privacy remain, since mobile devices have collected enormous amounts of personal data such as biometric data, photos, and other personal information, and sent them to remote servers. Personalized recommendation service, as an example, is used to manage users flight, meeting, and hotel booking information and provide recommendations based on users personal information such as contacts, message, calendar, location, sports/sleep data, app usage, etc. In this case, horizontal FML techniques provide a secure and trustworthy solution to prevent leakage of sensitive personal data. 3.3. HEALTHCARE There are diverse health-related data such as trans-omics data, including genome, epigenome, transcriptome, metabolome, proteome and metagenome, imaging data and phenotype data collected from wearable devices or other channels, along with the environmental, socioeconomic and behavior data. However, health-related data, especially patients data is highly sensitive and distributed in nature, thus collection and sharing of such data may bring critical legal and ethical privacy concerns. For example, if insurers learn a patient’s health data and find out he/she has severe or high medical cost diseases, they may refuse to provide insurance service. FML can overcome those obstacles by providing a federated machine learning model across organizations while keeping sensitive health data in the local environment. FML applications in the healthcare field may have different scenarios including business-to-government (B2G), business-to-business (BB, business-to-customer (BC) or mixed models. The most common FML scenario in healthcare is BB, where there is a need for the collaborative building of FML models among different hospitals, companies, research institutions, etc. Direct moving data between hospitals may raise concerns about security, privacy, and availability of medical data. FML can address these concerns and the horizontal FML model should achieve better performance than the models trained with single institutional data. As an example, with horizontal FML, in genetic studies, the comprehensive analysis of genes helps to discover the hidden patterns between genotype and phenotype and benefits diagnostic and treatment development of diseases such as cancers. Currently, samples collected from a single institution is insufficient to coverall the mutations in BRCA1/2, while FML provides a feasible and secured way of training an FML model predicting the risk of breast and ovarian cancer. 3.4. EDUCATION Uses of machine learning in education and training applications range from standard data mining for the purpose of domain specific student assessment (such as language skill diagnosis, personalized learning, teacher’s aids, human knowledge discovery and representation, etc. The educational AI employs a variety of traditional machine Authorized licensed use limited to University of Malta. Downloaded on December 24,2022 at 11:03:39 UTC from IEEE Xplore. Restrictions apply. |