Triton - Transprecise Edge Computing
Triton - Transprecise Edge Computing
Disciplines
Other Technical Sciences (15%); Electrical Engineering, Electronics, Information Engineering (10%); Computer Sciences (75%)
Keywords
-
Edge Computing,
Sustainable ICT,
Distributed Machine Learning,
Transprecision
Digital transformation causes disruptive changes in our lives. Tools like ChatGPT and other AI tools demonstrate the possibility for life and work changing applications ranging from prompt-based generation of texts, images and videos to personalized learning or personalized medicine, and earth science utilized to prevent or to understand climate change. All those applications have in common the huge amount of data produced and the enormous energy consumption necessary for data storage, data processing, data transmission or to visualize data. For the increasing amount, of applications decisions have to be made within a fraction of a second, for example in the case of traffic management or virtual reality where huge amounts of data in different formats have to be processed in the vicinity of the user. End devices like virtual reality oculus are not powerful enough. As a consequence, we need powerful computational facilities in the vicinity of the users, at the traffic light or at the oculus placed at our head. As a solution, so-called micro data centers or edge nodes are computers placed outside of data centers delivering computation to end devices that cannot cover the computational demands. Currently computation on the edge is done in the best effort manner without any guarantees on quality or efficiency which might create issues especially in case of time critical applications like traffic management. In this project we develop novel fundamental methods and tools for measurable, provable, and programmable energy efficiency on the Edge called transprecision that can be used during the software engineering process (e.g., in the form of an algorithm). The methods for transprecision developed in this project will be demonstrated in concrete use cases. Our first use case is Augmented Reality (AR) that has very strict latency requirements but might experience varying workloads and has classical behavior patterns like mobility. Transprecision has the potential to achieve a higher utilization and performance at a lower cost, e.g., when only the relevant parts of the video are downloaded to nearby edge nodes. Our second use case is the smart traffic light in combination with 5G/6G equipped with cameras continuously observing specific areas for object detection. Transprecision has the potential to achieve relatively high accuracy of object detection by reducing the amount of processed data. In addition, we will demonstrate our concept of transprecision on the edge using hybrid architectures. Apparently, we are experiencing a new era of computer systems where part of the computation is done on the classic machine but specific parts are done on an analog machine, like quantum computers achieving an increase of efficiency by several orders of magnitude. Connecting both worlds (classic and quantum) is a tedious and complicated task that requires special connection of devices and computational facilities where edge computers play a crucial role.
- Technische Universität Wien - 100%
- Thomas Monz, Universität Innsbruck , national collaboration partner
Research Output
- 4 Publications
-
2024
Title FLIGAN DOI 10.1145/3642968.3654813 Type Conference Proceeding Abstract Author Maliakel P Pages 1-6 -
2024
Title The computing continuum: From IoT to the cloud DOI 10.1016/j.iot.2024.101272 Type Journal Article Author Al-Dulaimy A Journal Internet of Things Pages 101272 Link Publication -
2024
Title Machine Learning Workflows in the Computing Continuum for Environmental Monitoring DOI 10.1007/978-3-031-63775-9_27 Type Book Chapter Author Catalfamo A Publisher Springer Nature Pages 368-382 -
2023
Title SymED: Adaptive and Online Symbolic Representation of Data on the Edge DOI 10.48550/arxiv.2309.03014 Type Preprint Author Hofstätter D