Ieee federated Machine Learning White Paper
Download 0.56 Mb. View original pdf
|
e-Iraq estra.ar.en
- Navigate this page:
- 3.5. URBAN COMPUTING
| 13 IEEE SA determined a priori, but the input to the rules is often machine learned. For example, the rules might dictate different instructional behaviors based on the categorization of learners, and the categorizations might be learned from data. Or, the rules might require estimates of the effectiveness of learner activity, or classification of learning activity in educational taxonomies (e.g., Blooms. To the extent that learners or activities are exchanged by multiple systems, FML can be used to improve accuracy without compromising the privacy of data. In addition, it is possible that the rules themselves are machine learned. In that case, using methods like reinforcement learning, it is important to have large data sets by aggregating data from multiple sources across multiple students. 3.5. URBAN COMPUTING Urban computing is a process of acquiring, integrating, and analyzing heterogeneous big data generated by a diversity of sources including sensors, devices, vehicles, buildings, and humans in urban space. Applications of urban computing aim to tackle major city challenges, such as air pollution, increased energy consumption, and traffic congestion. Urban computing also helps us understand the nature of urban phenomena and even predict the future of cities. Urban computing is an interdisciplinary field fusing the computing science field with traditional fields like transportation, civil engineering, economy, ecology, and sociology in the context of urban spaces. In many urban computing scenarios, there are regulatory requirements and privacy concerns that prevent the sharing of data. FML can overcome these challenges by building a federated machine learning model across organizations while the individual data of each organization stay in their local environment. Smart ride-hailing is one example of this use case. Ride-hailing companies have a strong incentive to find optimal solutions to the Vehicle Routing Problem. GPS data from these vehicles provide information on the number of vehicles and their speed along with different road segments, facilitating the predictions about future traffic conditions. While ride- hailing companies are not allowed or maybe unwilling to share valuable data with each other, FML solves the problem by allowing different companies to build and train a federated machine learning model, with model parametersbut not the private datasecurely exchanged under the federated system’s encryption mechanism. Environmental protection is another example of the use case. Air quality prediction can help residents take precautionary measures and allow city governments to implement corresponding countermeasures. However, this can be challenging since the air quality of a region depends on many factors, including industrial emissions, vehicle exhaust, and meteorological conditions. Factories are not allowed or may not be willing to share data about their real-time emissions, from which sensitive operational and financial information maybe gleaned. Regulatory and privacy concerns may also prevent environmental regulators from collecting data about individual vehicles location, model, and speed, air quality index (AQI) readings from sparsely distributed air quality Authorized licensed use limited to University of Malta. Downloaded on December 24,2022 at 11:03:39 UTC from IEEE Xplore. Restrictions apply. |