About this Research Topic
During the last decade, advances in artificial intelligence (AI) such as convolutional neural networks, deep learning, and hardware accelerators have enabled the widespread application of machine learning (ML) in many areas, e.g. as computer vision and natural language processing, etc. This trend is expected to continue and expand in the coming years, leading to a world that depends heavily on ML-based systems. In particular, when ML is applied for safety-critical scenarios, the reliability of the ML system becomes a more important problem. Space application is such an scenario where cosmic radiation poses a severe threat to the dependability of ML system.
AI continues to have a major role in enabling future smart cities, surveillance, smart sensing, and so on, and will also play an important role for smart satellite or smart space networks. For space applications, the safety and reliability of the AI system is a more critical requirement. This poses many research challenges. For example, fault tolerance is commonly achieved by redundant design, but the overhead of deep neural networks would impose a big burden to the resource-limited space platform. In addition, space radiation will introduce soft or hard errors on ML accelerators, especially for FPGA based implementation. Understanding the effects of soft errors on advanced ML systems is a complex issue, and is an important precondition for efficient fault tolerance design. Based on above concerns, this Research Topic is devoted to: 1) reliability evaluation and analysis for advanced ML algorithms and accelerators; 2) efficient fault tolerance design to improve the reliability of ML systems.
Potential topics include, but are not limited to, the following:
- Reliability evaluation of ML for safety-critical applications
- Reliability comparison of different ML accelerators
- Efficient fault detection of typical ML algorithms
- Efficient fault/error-tolerant ML systems and techniques
- Reliability-oriented ML algorithm design
- Efficient architecture search for reliable neural networks
- New system structures for reliable AI
Keywords: Machine Learning, Neural Networks, Reliability, Evaluation, Fault Tolerance, Space Platform
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.