Ieee federated Machine Learning White Paper
Download 0.56 Mb. View original pdf
|
e-Iraq estra.ar.en
- Navigate this page:
- 3.9. IOT/EDGE COMPUTING
| 15 IEEE SA an accurate statistical marketing model. This eventually restricts the practical performance of a marketing strategy. As the increasingly strict data privacy regulations take effect, including the GDPR, and the public awareness of data confidentiality keeps rising, the sharing of data among various subareas is generally prohibited subject to data privacy concerns, which in turn impedes the development of high-performance models for traditional marketing platforms. The introduction of FML, especially vertical FML across different organizations that complement each other in data dimensions, conducts virtual model aggregation without risk of privacy breaching when making decisions. A wide range of features and samples achieved by FML help enrich useful patterns that can be extracted for training a machine learning model and thereby significantly improve the marketing model performance. Cooperating with client's social behavior data collected from social media companies via FML, for example, credit rating companies have the capability to identify clients with potentially high default rates, which can only be computed by collectively checking on multiple financial organizations. 3.9. IOT/EDGE COMPUTING With the development of G and Internet of Things (IoT) technology, edge devices, and local data are fast growing. One of the main trends in G and edge computing is to power the edge devices with intelligence instead of leaving the intelligence in a server cloud. As a result, more mobile phones and IoT devices require the ability to make decisions locally while collaborating with servers globally. The ability to process data and build machine learning models locally to enable privacy-preservation as well as personalize user experience is an important area across the academic and industrial domains. Traditional AI performs the cloud-based or centralized modeling process, which needs data transferred to the cloud servers and conducting the training, evaluation, deployment, and serving. With the introduction of federated learning, collaborative training can be widely used on mobile phones or IoT devices to enable data and model enhancement. AI scenarios on edge devise can be improved, which includes personalized image processing for cameras, short video content generation, better automatic speech recognition (ASR) and natural language understanding (NLU) for personal virtual assistants, improved recommendation system, online advertising, smart air-conditioner and intelligent glasses with AR/VR, etc. Most of the existing pipelines of aggregating local data into a logically or physically centralized server for processing can be extended with FML. As AI applications are getting increasingly popular on mobile devices, the application developers are no longer satisfied with AI models being trained with opened datasets. They also wish to collect information from users to optimize their models for improving model performance and user experience. However, collecting personal data Authorized licensed use limited to University of Malta. Downloaded on December 24,2022 at 11:03:39 UTC from IEEE Xplore. Restrictions apply. |