Current Issue
lssue 7, 2025
Editors-in-Chief
Yunhe Pan, Aiguo Fei
Advisor Xicheng Lu
ISSN 2095-9184
CN 33-1389/TP
IF (2024 JCR) 2.9
| Submit Now |
- Current Issue
- Cover Articles
- Archive
- Virtual Issues
- Online First
Volume 26 Issue 7，2025
Regular Papers
Towards the first principles of explaining DNNs: interactions explain the learning dynamics AI Introduction
“In the field of explainable artificial intelligence, this paper discusses whether interaction-based explanation can serve as the first-principles explanation of a deep neural network. Expert established the interaction theory system, which provides solutions to solve the extremely complex learning dynamics of a DNN.”Abstract：Most explanation methods are designed in an empirical manner, so exploring whether there exists a first-principles explanation of a deep neural network (DNN) becomes the next core scientific problem in explainable artificial intelligence (XAI). Although it is still an open problem, in this paper, we discuss whether the interaction-based explanation can serve as the first-principles explanation of a DNN. The strong explanatory power of interaction theory comes from the following aspects: (1) it establishes a new axiomatic system to quantify the decision-making logic of a DNN into a set of symbolic interaction concepts; (2) it simultaneously explains various deep learning phenomena, such as generalization power, adversarial sensitivity, representation bottleneck, and learning dynamics; (3) it provides mathematical tools that uniformly explain the mechanisms of various empirical attribution methods and empirical adversarial-transferability-boosting methods; (4) it explains the extremely complex learning dynamics of a DNN by analyzing the two-phase dynamics of interaction complexity, which further reveals the internal mechanism of why and how the generalization power/adversarial sensitivity of a DNN changes during the learning process.Keywords：First-principles explanation;Theory of equivalent interactions;Two-phase dynamics of interactions;Learning dynamics57|27|0Updated：2025-07-28Regular Papers
Image generation evaluation: a comprehensive survey of human and automatic evaluations AI Introduction
“In the field of image generation, this paper comprehensively surveys the research progress of image evaluation, including evaluation protocols and methods. It proposes a novel protocol covering human and automatic evaluation aspects for various image generation tasks and reviews automatic evaluation methods in the past five years. The paper lays a foundation for the construction of image generation evaluation systems.”Abstract：Image generation models have made remarkable progress, and image evaluation is crucial for explaining and driving the development of these models. Previous studies have extensively explored human and automatic evaluations of image generation. Herein, these studies are comprehensively surveyed, specifically for two main parts: evaluation protocols and evaluation methods. First, 10 image generation tasks are summarized with focus on their differences in evaluation aspects. Based on this, a novel protocol is proposed to cover human and automatic evaluation aspects required for various image generation tasks. Second, the review of automatic evaluation methods in the past five years is highlighted. To our knowledge, this paper presents the first comprehensive summary of human evaluation, encompassing evaluation methods, tools, details, and data analysis methods. Finally, the challenges and potential directions for image generation evaluation are discussed. We hope that this survey will help researchers develop a systematic understanding of image generation evaluation, stay updated with the latest advancements in the field, and encourage further research.Keywords：Image generation evaluation;Human evaluation;Automatic evaluation;Evaluation protocols;Evaluation aspects57|41|0Updated：2025-07-28Regular Papers
Abstract：The task of recognizing Chinese variant characters aims to address the challenges of semantic ambiguity and confusion, which potentially cause risks to the security of Web content and complicate the governance of sensitive words. Most existing approaches predominantly prioritize the acquisition of contextual knowledge from Chinese corpora and vocabularies during pretraining, often overlooking the inherent phonological and morphological characteristics of the Chinese language. To address these issues, we propose a shared-weight multimodal translation model (SMTM) based on multimodal information of Chinese characters, which integrates the phonology of Pinyin and the morphology of fonts into each Chinese character token to learn the deeper semantics of variant text. Specifically, we encode the Pinyin features of Chinese characters using the embedding layer, and the font features of Chinese characters are extracted based on convolutional neural networks directly. Considering the multimodal similarity between the source and target sentences of the Chinese variant-character-recognition task, we design the shared-weight embedding mechanism to generate target sentences using the heuristic information from the source sentences in the training process. The simulation results show that our proposed SMTM achieves remarkable performance of 89.550% and 79.480% on bilingual evaluation understudy (BLEU) and F1 metrics respectively, with significant improvement compared with state-of-the-art baseline models.Keywords：Chinese variant characters;Multimodal model;Translation model;Phonology and morphology30|31|0Updated：2025-07-28Regular Papers
“In the field of on-orbit low-light image enhancement, this study introduces its research progress. Expert xx established the BeiDou navigation satellite dataset, which provides solutions to solve on-orbit low-light image enhancement problems.”Abstract：On-orbit service is important for maintaining the sustainability of the space environment. A space-based visible camera is an economical and lightweight sensor for situational awareness during on-orbit service. However, it can be easily affected by the low illumination environment. Recently, deep learning has achieved remarkable success in image enhancement of natural images, but it is seldom applied in space due to the data bottleneck. In this study, we first propose a dataset of BeiDou navigation satellites for on-orbit low-light image enhancement (LLIE). In the automatic data collection scheme, we focus on reducing the domain gap and improving the diversity of the dataset. We collect hardware-in-the-loop images based on a robotic simulation testbed imitating space lighting conditions. To evenly sample poses of different orientations and distances without collision, we propose a collision-free workspace and pose-stratified sampling. Subsequently, we develop a novel diffusion model. To enhance the image contrast without over-exposure and blurred details, we design fused attention guidance to highlight the structure and the dark region. Finally, a comparison of our method with previous methods indicates that our method has better on-orbit LLIE performance.Keywords：Satellite capture;Low-light image enhancement (LLIE);Data collection;Diffusion model;Fused attention28|26|0Updated：2025-07-28Regular Papers
An end-to-end automatic methodology to accelerate the accuracy evaluation of deep neural networks under hardware transient faults AI Introduction
“In the field of deep neural networks (DNNs), hardware transient faults significantly impact safety-critical applications. Expert researchers have designed an automatic methodology, A-Mean, which uses silent data corruption rate and static two-level mean calculation to rapidly estimate classification accuracy and safety-critical misclassification. This easy-to-use tool achieves up to 922.80 times speedup compared to state-of-the-art methods, with minimal loss in safety and accuracy.”Abstract：Hardware transient faults are proven to have a significant impact on deep neural networks (DNNs), whose safety-critical misclassification (SCM) in autonomous vehicles, healthcare, and space applications is increased up to four times. However, the inaccuracy evaluation using accurate fault injection is time-consuming and requires several hours and even a couple of days on a complete simulation platform. To accelerate the evaluation of hardware transient faults on DNNs, we design a unified and end-to-end automatic methodology, A-Mean, using the silent data corruption (SDC) rate of basic operations (such as convolution, addition, multiply, ReLU, and max-pooling) and a static two-level mean calculation mechanism to rapidly compute the overall SDC rate, for estimating the general classification metric accuracy and application-specific metric SCM. More importantly, a max-policy is used to determine the SDC boundary of non-sequential structures in DNNs. Then, the worst-case scheme is used to further calculate the enlarged SCM and halved accuracy under transient faults, via merging the static results of SDC with the original data from one-time dynamic fault-free execution. Furthermore, all of the steps mentioned above have been implemented automatically, so that this easy-to-use automatic tool can be employed for prompt evaluation of transient faults on diverse DNNs. Meanwhile, a novel metric "fault sensitivity" is defined to characterize the variation of transient fault-induced higher SCM and lower accuracy. The comparative results with a state-of-the-art fault injection method TensorFI+ on five DNN models and four datasets show that our proposed estimation method A-Mean achieves up to 922.80 times speedup, with just 4.20% SCM loss and 0.77% accuracy loss on average. The artifact of A-Mean is publicly available at https://github.com/breatrice321/A-Meanhttps://github.com/breatrice321/A-Mean.Keywords：Analytical model;Deep neural networks;Hardware transient faults;Fast evaluation;Automatic evaluation tool27|34|0Updated：2025-07-28Regular Papers
CRGT-SA: an interlaced and spatiotemporal deep learning model for network intrusion detection AI Introduction
“In the field of cybersecurity, a novel interlaced and spatiotemporal deep learning model called CRGT-SA has been proposed. This model combines CNN with gated TCN and RNN modules to learn spatiotemporal properties and imports the self-attention mechanism to select significant features, providing solutions to protect the Internet from complicated cyberattacks.”Abstract：To address the challenge of cyberattacks, intrusion detection systems (IDSs) are introduced to recognize intrusions and protect computer networks. Among all these IDSs, conventional machine learning methods rely on shallow learning and have unsatisfactory performance. Unlike machine learning methods, deep learning methods are the mainstream methods because of their capability to handle mass data without prior knowledge of specific domain expertise. Concerning deep learning, long short-term memory (LSTM) and temporal convolutional networks (TCNs) can be used to extract temporal features from different angles, while convolutional neural networks (CNNs) are valuable for learning spatial properties. Based on the above, this paper proposes a novel interlaced and spatiotemporal deep learning model called CRGT-SA, which combines CNN with gated TCN and recurrent neural network (RNN) modules to learn spatiotemporal properties, and imports the self-attention mechanism to select significant features. More specifically, our proposed model splits the feature extraction into multiple steps with a gradually increasing granularity, and executes each step with a combined CNN, LSTM, and gated TCN module. Our proposed CRGT-SA model is validated using the UNSW-NB15 dataset and is compared with other compelling techniques, including traditional machine learning and deep learning models as well as state-of-the-art deep learning models. According to the simulation results, our proposed model exhibits the highest accuracy and F1-score among all the compared methods. More specifically, our proposed model achieves 91.5% and 90.5% accuracy for binary and multi-class classifications respectively, and demonstrates its ability to protect the Internet from complicated cyberattacks. Moreover, we conduct another series of simulations on the NSL-KDD dataset; the simulation results of comparison with other models further prove the generalization ability of our proposed model.Keywords：Intrusion detection;Deep learning;Convolutional neural network;Long short-term memory;Temporal convolutional network26|22|0Updated：2025-07-28
SEE MORE
0 1 Hybrid-augmented intelligence: collaboration and cognition
0 3 Applications of artificial intelligence in intelligent manufacturing: a review*
0 5 Challenges and opportunities: from big data to knowledge in AI 2.0
0 6 Recent advances in efficient computation of deep convolutional neural networks
0 7 Terahertz time-domain spectroscopy and micro-cavity components for probing samples: a review
0 8 Cross-media analysis and reasoning: advances and directions
0 9 Current trends in the development of intelligent unmanned autonomous systems
SEE MORE
Videos
-
00:02:512023 Issue 1 | Scalability and efficiency challenges for the exascale supercomputing system: practice of a parallel supporting environment on the Sunway exascale prototype system
2023-12-30Play Total： 23 - 00:02:30
2023 Issue 6 | Model division multiple access for semantic communications
2023-12-30Play Total： 13 - 00:02:15
2022 Issue 10 | Discussion on a new paradigm of endogenous security towards 6G networks
2023-12-30Play Total： 0 - 00:02:22
2022 Issue 12 | Technology trends in large-scale high-efficiency network computing
2023-12-30Play Total： 2 - 00:02:48
2022 Issue 6 | Self-deployed execution environment for high performance computing
2022-08-03Play Total： 7 - 00:02:24
2022 Issue 2 | A full-process intelligent trial system for smart court
2022-05-17Play Total： 8 - 00:02:37
2022 Issue 3 | Automatic protocol reverse engineering for industrial control systems with dynamic taint analysis
2022-05-17Play Total： 5 - 00:05:36
P1 Speech by Academician Baoyan Duan
2022-04-17Play Total： 10 - 00:02:27
P2 Speech by Professor Min Sheng, Xidian University
2022-04-17Play Total： 6 - 00:02:37
P3 Speech by Professor Yunsong Li, Xidian University
2022-04-17Play Total： 11
SEE MORE
- Address：Zhejiang University Press, 148 Tianmushan Road, Hangzhou, China Postal code：310028
- Tel：+86-571-88273162 Email：fitee@zju.edu.cn
- Copyright © 2000 - 2025, Zhejiang University Press 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0