Machine Learning in Resource-Constrained Embedded Systems
摘要截稿:
全文截稿: 2020-07-25
影响因子: 1.161
期刊难度:
CCF分类: 无
中科院JCR分区:
• 大类 : 计算机科学 - 4区
• 小类 : 计算机:硬件 - 4区
• 小类 : 计算机:理论方法 - 4区
• 小类 : 工程:电子与电气 - 4区
Overview
Machine learning (ML) methods, in the presence of deep learning techniques and big data gathered by the emergence of Internet of Things (IoT) and Cyber-Physical Systems (CPS), play a critical role in extracting meaningful information from the surrounding world. Transferring such amount of data to data-centers and clouds for storage and computational goals, either for training or for inference, may not be always possible because of costs like time, energy, network, and so on. These costs may not be acceptable for many real-world applications, including time-sensitive, battery-operated, and connectivity-limited devices. A resulting trend is thus, in-sensor/near-sensor computations and/or domain-specific architectures which perform optimizations using specialized ML accelerators. On the other hand, ML applications often need to achieve high utility/accuracy under certain resource constraints. The constraints may be changed to optimization goals as well, depending on the application type. Addressing this tradeoff is an inherent challenge that needs to be investigated in a principled fashion in order to understand the physical world more practically and effectively, raising the need to upgrade and adapt ML algorithms. In this special issue, we welcome original submissions in all theoretical and practical, yet application-specific, design and analysis methods of machine learning, deep learning, and artificial intelligence in the context of embedded systems and edge-computing.
Topics of interest include (but not limited to):
Design methodologies for resource-constrained ML hardware accelerators
ML workload acceleration on conventional technologies like FPGA, ASIP and ASIC
Acceleration of domain-specific ML algorithms for edge-computing, IoT, and CPS
Approximate ML for resource-constrained systems
Resource-aware approximate ML
Design methodologies for scalable ML
Design methodologies for adaptive ML
Metrics and methods to evaluate ML accuracy/utility