Call for Paper on Special Issue on Explainable Deep Learning for Efficient and Robust Pattern Recognition
• 大类 : 工程技术 - 2区
• 小类 : 计算机：人工智能 - 2区
• 小类 : 工程：电子与电气 - 2区
Deep neural networks (DNNs) have recently achieved outstanding predictive performance, and become an indispensable tool in a wide range of pattern recognition applications, including image classification, object detection, video understanding, document analysis, etc. While DNN methods give impressively high predictive accuracy, they are often perceived as black-boxes with deep, computationally expensive layers, and have been recently found vulnerable to spoofing with well-designed input samples in many safety critical applications. This is especially so in several sensitive or real-time pattern recognition applications including medical diagnosis, face recognition and self-driving cars. In these applications, a single incorrect prediction might be very costly, and thus the reliance on the trained model and its capacity to deliver both efficient and robust data processing must be guaranteed. Therefore, understanding how the DNN behaves, and thus generating explainable deep learning models have become an essential and fundamental problem.
Even with the recent extraordinary rapid progress and significant successes of DNNs over a wide spectrum of applications, it is unfortunately not clear what information must be present in the input data and how it must be used in deep learning models to guarantee fast, safe and stable prediction. There has been an explosion of interest recently along related research directions, such as a) analysing the information bottleneck for efficient learning, b) inferring and regularizing the network structure for stable and robust prediction, and c) interpreting the learned representations and generated decisions. This encouraging progress brings profound implications for research into the topic of this special edition, namelyexplainable deep learning for efficient and robust pattern recognition. This important and open research topic brings new challenges and opportunities to the pattern recognition community. It requires the application of tremendous efforts to uncover the fundamental mechanisms from several different point of views, including information theory, machine learning, computer vision, information security, etc. Moreover, it also potentially benefits a variety of closely related areas in pattern recognition areas, and opens up the possibility of practical safety-critical or low-cost applications.
The goal of this special issue is to broadly engage the different communities together and provide a forum for the researchers and practitioners related to this rapidly developed field to share their novel and original research of the topicexplainable deep learning for efficient and robust pattern recognition. Survey papers addressing relevant topics are also welcome.
The page limit for regular papers is 35 single column double spaced pages, and 40 pages for reviews. Authors should clearly state the relationship between the submitted paper, and any related previously published conference or journal papers. Submissions to the special edition must contain a substantial amount of original previously unpublished material. Topics of interest include, but are not limited to:
•Novel theoretical insights on the deep neural networks
•Quantifying or visualizing the interpretability of deep neural networks
•Stability improvements of deep neural network optimization
•Deep network compression/acceleration in pattern recognition applications
•Compact architecture and the regularization of deep neural networks
•Adversarial attacks and defending critical applications in pattern recognition
•Deep neural network in safety-critical or low-cost pattern recognition