Cyclic Features Research Articles

Dynamic Structural Equation Modeling with Cycles

Cyclical phenomena are commonly observed in many areas of repeated measurements, especially with intensive longitudinal data. A typical example is circadian (24-hour) rhythm of physical measures such as blood pressure, heart rate, glucose level, and alertness. This paper focuses on positive affect, which is a common measure in psychological studies and for which circadian rhythm has been observed but not analyzed by modern statistical methods. The paper demonstrates that a large new analysis arsenal is available for analysis of cyclical features in intensive longitudinal data. This can help researchers extract more information from their data to get more valid estimates of coupled processes and to get new theoretical insights into circadian rhythms of mood. To assist in this effort, the analyses are based on general models with a rich set of features while still being accessible without an unduly steep learning curve. Scripts for the Mplus software are available for all the analyses presented.

Psychological Correlates of Post-COVID Condition

<span lang="SK">This article deals with psychological correlates of the post-COVID condition - depression, anxiety, and memory problems including remembering and forgetting. The work has a longitudinal character, while it examines the post-COVID condition in a group of participants based on infection of the disease Covid-19. At the same time, participants who never had this disease (N=86, for the 1. measure and N=38 for the 2. measure) are compared with those who were infected with COVID-19 (N=112, for the 1. measure and N=36 for the 2. measure), while participants were separated for analysis in groups based on the time from infection. Group of respondents who were infected within 3 months, from 3-9 months from infection and 9+ months since infecting with the disease. We used tests for comparison, such as the Kruskal Wallis test and multivariate procedures for within-subject and between-subject changes, using Anova for mixed experimental designs. We also determined the risk of being included in a group based on the time of infection with COVID-19 through multinominal logistic regression. We found differences in all variables in general between those infected with COVID-19 and those not. In neuropsychiatric aspects- anxiety and depression were confirmed cyclical features in those who were infected with COVID-19 in time, with the worst value of variables in the group from 3-9 months from infecting with the disease. In the current memory scale and its subscale, the act of remembering was a value that worsened over time and it was confirmed as an effect of group inclusion.</span>

ADMNet: An adaptive downsampling multi-frequency multi-channel network for long-term time series forecasting

Long-term time series forecasting finds widespread applications in various domains such as energy, finance, and transportation. Decomposing time series into sub-sequences with distinct temporal relationships (periods) for analysis and modeling is an effective approach for long-term time series prediction. However, conventional decomposition techniques often rely on fixed-size parameter kernels for sliding averages, leading to inaccurate and unreasonable captures of the underlying periodicities in the time series. Additionally, it is essential to acknowledge that time series data often exhibits multi-periodic characteristics. In many time series prediction tasks, the value at a future time point is influenced by multiple periods within the time series. To accurately and flexibly capture the periodicities of time series, we propose an enhanced adaptive cyclic feature recognizer to automatically identify the periodic lengths for sliding average parameter kernel sizing. To fully exploit the multiple periodicities inherent in time series, we merge and smooth the sequences of features corresponding to the identified cycles, obtaining two frequency-blended cyclic feature fusion terms. Furthermore, we extract high-frequency random components to preserve finer details. Finally, we individually model the three cyclic feature fusion terms. In summary, we introduce the Adaptive Multi-Frequency Multi-Channel Network (ADMNet) designed to autonomously capture multi-periodic features. Experimental results show that, compared to the current state-of-the-art seven benchmark models, the proposed model achieves an average reduction in mean squared error (MSE) ranging from 6.19% to 22.75% across eight datasets, including ETT, Traffic, and Weather. This indicates that our model delivers superior predictive performance on multiple real-world datasets.

A review of multiaxial low-cycle fatigue criteria for life prediction of metals

Most of real-world structural components that undergo cyclic loading feature multiaxial fatigue. When the cyclic loading involves also significant plastic deformation, multiaxial low-cycle fatigue takes place. Applications where multiaxial low-cycle fatigue can be observed very often involve metal components. To predict their lives multiple criteria and models have been proposed, but their development has not followed a regular path. Multiple reviews are available in literature. However, many of them are outdated, they often employ different classification methods to categorize available criteria, many focus on specific families of criteria, and others do not include sufficient theoretical background. Moreover, none of the available reviews is based on a systematic literature search method. As a result, approaching the topic can result arduous and chaotic, especially for first timers. This work aims at providing a clear, comprehensive, and definitive review of available criteria for multiaxial low-cycle fatigue. First, the basic theoretical background is explained. Secondly, a systematic approach is described and employed to identify all major currently available criteria. Then, they are classified and commentary about different classification styles that can be found in literature is added. Eventually they are described, together with their latest proposed variations. In this way this review can be employed as a guiding reference, especially for engineers approaching the topic for the first time.

Expression patterns of folate metabolism-related enzymes in the bovine oviduct: estrous cycle-dependent modulation and responsiveness to folic acid

Folate metabolism is required for important biochemical processes that regulate cell functioning, but its role in female reproductive physiology in cattle during peri- and post-conceptional periods has not been thoroughly explored. Previous studies have shown the presence of folate in bovine oviductal fluid, as well as finely regulated gene expression of folate receptors and transporters in bovine oviduct epithelial cells (BOECs). Additionally, extracellular folic acid (FA) affects the transcriptional levels of genes important for the functioning of BOECs. However, it remains unknown whether the anatomical and cyclic features inherent to the oviduct affect regulation of folate metabolism. The present study aimed to characterize the gene expression pattern of folate cycle enzymes in BOECs from different anatomical regions during the estrous cycle and to determine the transcriptional response of these genes to increasing concentrations of exogenous FA. A first PCR screening showed the presence of transcripts encoding dihydrofolate reductase (DHFR), methylenetetrahydrofolate reductase (MTHFR), and methionine synthase (MTR) in bovine reproductive tissues (ovary, oviduct and uterus), with expression levels in oviductal tissues comparable to, or even higher than, those detected in ovarian and uterine tissues. Moreover, expression analysis through RT-qPCR in BOECs from the ampulla and isthmus during different stages of the estrous cycle demonstrated that folate metabolism-related enzymes exhibited cycle-dependent variations. In both anatomical regions, DHFR was upregulated during the preovulatory stage, while MTHFR and MTR exhibited increased expression levels during the postovulatory stage. Under in vitro culture conditions, ampullary and isthmic cells were cultured in the presence of 10, 50, and 100 μM FA for 24 h. Under these conditions, isthmus epithelial cells exhibited a unique transcriptional response to exogenous FA, showing a pronounced increase in MTR expression levels. Our results suggest that the expression of folate metabolism-related genes in BOECs is differentially regulated during the estrous cycle and may respond to exogenous levels of folate. This offers a new perspective on the transcriptional regulation of genes associated with the folate cycle in oviductal cells and provides groundwork for future studies on their functional and epigenetic implications within the oviductal microenvironment.

Revisiting the Pasinetti Index: Understanding Its Cyclical and Long-Term Features and Its Important Implications for Macroeconomic Policy

ABSTRACT This article explores the long-term and cyclical effects of the Pasinetti Index (PI) and their implications for income distribution and macroeconomic policy in a post-pandemic environment marked by inflationary pressures and increasingly restrictive central bank policies. It discusses the relevance of adopting a zero PI as monetary policy framework as either a short-term target or a long-term norm. The research underscores the importance of coordinating monetary and fiscal policies to achieve a balanced mix of short-term and long-term macroeconomic goals, as per the Pasinetti rule, aimed at stabilizing income distribution between rentier and non-rentier groups without compromising on a Keynesian full employment commitment. By analyzing historical data and employing a SVAR model for Canada and the United States, the study highlights the significant cyclical impact of PI fluctuations on unemployment and income distribution. The findings challenge the efficacy of rigid monetary policy rules and support a macroeconomic policy that reconciles short-term employment objectives and long-term distributional goals.

New BDNF and NT-3 Cyclic Mimetics Concur with Copper to Activate Trophic Signaling Pathways as Potential Molecular Entities to Protect Old Brains from Neurodegeneration.

A low level of Neurotrophins (NTs), their Tyrosine Kinase Receptors (Trks), Vascular Endothelial Growth Factors (VEGFs) and their receptors, mainly VEGFR1 and VEGFR2, characterizes AD brains. The use of NTs and VEGFs as drugs presents different issues due to their low permeability of the blood-brain barrier, the poor pharmacokinetic profile, and the relevant side effects. To overcome these issues, different functional and structural NT mimics have been employed. Being aware that the N-terminus domain as the key domain of NTs for the binding selectivity and activation of Trks and the need to avoid or delay proteolysis, we herein report on the mimicking ability of two cyclic peptide encompassing the N-terminus of Brain Derived Growth Factor (BDNF), (c-[HSDPARRGELSV-]), cBDNF(1-12) and of Neurotrophin3 (NT3), (c-[YAEHKSHRGEYSV-]), cNT3(1-13). The two cyclic peptide features were characterized by a combined thermodynamic and spectroscopic approach (potentiometry, NMR, UV-vis and CD) that was extended to their copper(II) ion complexes. SH-SY5Y cell assays show that the Cu2+ present at the sub-micromolar level in the complete culture media affects the treatments with the two peptides. cBDNF(1-12) and cNT3(1-13) act as ionophores, induce neuronal differentiation and promote Trks and CREB phosphorylation in a copper dependent manner. Consistently, both peptide and Cu2+ stimulate BDNF and VEGF expression as well as VEGF release; cBDNF(1-12) and cNT3(1-13) induce the expression of Trks and VEGFRs.

Comparative Analysis Of Cycle-To-Cycle Variabilities And Combustion Development In An Optical Spark-Ignition Engine Fueled By Pure Hydrogen And Propane: Insights From Chemiluminescence and PI

Hydrogen, derived from renewable sources and devoid of carbon emissions, is a pivotal energy carrier for the future. Representing a viable substitute for fossil fuels in internal combustion engines (ICEs), contemporary studies advocate using very-lean and ultra-lean hydrogen-air mixtures, with a fuel-air equivalence ratio below 0.5, as a potent strategy to reduce NOx emissions in hydrogen-fueled ICEs. An experimental setup with an optical spark-ignition engine was devised to investigate the primary factors influencing cycle-to-cycle variations in H2ICEs by optimizing mixture homogeneity and focusing the study on flame interaction with in-cylinder aerodynamics. Simultaneous in-cylinder pressure, chemiluminescence, and PIV analyses were performed to characterize and compare the behaviors of ultra-lean hydrogen-air and propane-air flames, assessing flame development and cyclic variation features. Results indicate that the flame development of hydrogen and propane is significantly different. Propane exhibited increased in-cylinder pressure trace variability at lean and rich flammability limits with a high flame development difference between fast and slow cycles. In contrast, hydrogen-air mixtures under various equivalence ratios (from very-lean to ultra-lean) presented much more stable in-cylinder pressures. Furthermore, fast propane flames were mainly advected to the cylinder's symmetrical axes, while fast hydrogen flames started further away from the spark plug. Horizontal and vertical PIV measurements showed a single structure flow field over the studied conditions with mainly intensity variations over the measured cycles. Fast flames were associated with more intense tumble motion. The differences between fast and slow flames for hydrogen are attributed to the early flame development. However, after initial differences in flame development over several crank angles, there are similarities in all cycles, suggesting that the low variability in in-cylinder pressures is mostly due to the robustness of hydrogen flame ignition and its independence from in-cylinder flow small variations at the early development phase.

Analysis of Model Development and Usage Patterns in Geological, Astronomical, and Atmospheric Domains in Korean Elementary School Science Textbooks

In science education, models are used as key teaching tools to teach scientific phenomena and concepts to students. Although South Korea recognizes the value of model use in science education, it is necessary to analyze how model use is reflected in science textbooks and in classrooms. This study examined the development and usage patterns of models in experimental activities in the geological, astronomical, and atmospheric domains in three Korean elementary school science textbooks. A coding framework was developed and analyzed based on previous studies and science curricula in other countries. The study found that, first, the model development and usage patterns in experimental activities in the geological, astronomical, and atmospheric domains in science textbooks were largely similar, although some differences existed across domains. Second, the resources required for model development and development methods; evidence-based model development, evaluation, and modification; and the cyclic features of modeling were similar across the three domains, but the geological domain was characterized by model limitations and the astronomical domain was characterized by the process of investigating a phenomenon. Based on these f indings, the study discussed model development activities in textbooks can be improved.

SCAN Algorithm to Optimize Dynamic Instructional Information Network for Predicting Student Behavior

With the integration of work, study and life on campus taking shape gradually in major universities across China, the concept of education from classroom to life is more widely accepted and the smart education model has become a trend in education informatization. However, most of the existing studies are not applicable to student behaviour data in the campus environment, and the temporal as well as cyclical characteristics in the data are not accessible in the student behaviour prediction problem. In this study, a hypergraph-based dynamic campus behaviour information network is designed to address the needs of the student behaviour prediction problem, and a student campus behaviour prediction algorithm is proposed in the dynamic campus behaviour information network-based student behaviour prediction problem. The effectiveness and rationality of the algorithm is verified through experiments with real campus data sets. The experimental results demonstrated that the periodic nature of the data acquired by the cycle gated cyclic unit module needs to be built on top of the snapshot gated cyclic unit module to help the algorithm achieve better results. The area under the curve of the algorithm proposed in the study achieves more advantageous results on both the 21-day and 35-day datasets. The cycle gated cyclic unit module of the student campus behaviour prediction algorithm proposed in this study can more effectively extract the cyclic features present in the dynamic campus behaviour information network and better accomplish the prediction of student campus behaviour.

A Machine Learning Model for Predicting Sleep and Wakefulness Based on Accelerometry, Skin Temperature and Contextual Information.

Body-worn accelerometers are commonly used to estimate sleep duration in population-based studies. However, since accelerometry-based sleep/wake-scoring relies on detecting body movements, the prediction of sleep duration remains a challenge. The aim was to develop and evaluate the performance of a machine learning (ML) model to predict accelerometry-based sleep duration and to explore if this prediction can be improved by adding skin temperature data, circadian rhythm based on the estimated midpoint of sleep, and cyclic time features to the model. Twenty-nine adults (17 females), mean (SD) age 40.2 (15.0) years (range 17-70) participated in the study. Overnight polysomnography (PSG) was recorded in a sleep laboratory or at home along with body movement by two accelerometers with an embedded skin temperature sensor (AX3, Axivity, UK) positioned at the low back and thigh. The PSG scoring of sleep/wake was used as ground truth for training the ML model. Based on pure accelerometer data input to the ML model, the specificity and sensitivity for predicting sleep/wake was 0.52 (SD 0.24) and 0.95 (SD 0.03), respectively. Adding skin temperature data and contextual information to the ML model improved the specificity to 0.72 (SD 0.20), while sensitivity remained unchanged at 0.95 (SD 0.05). Correspondingly, sleep overestimation was reduced from 54 min (228 min, limits of agreement range [LoAR]) to 19 min (154 min LoAR). An ML model can predict sleep/wake periods with excellent sensitivity and moderate specificity based on a dual-accelerometer set-up when adding skin temperature data and contextual information to the model.

United Cyclic Agree

The operation Agree has been defined in many different ways for different purposes. This presents a problem for minimalist syntax, since no one form of Agree can explain all of the relevant generalizations. It becomes especially problematic when multiple versions of Agree must exist simultaneously on the same probe in order to explain different phenomena. This article examines two such phenomena in Basque finite verb agreement: ergative displacement and the Strong Person-Case Constraint, and the theories that account for them: Cyclic Agree (Béjar and Rezac 2009) and Feature Gluttony (Coon and Keine 2021). A combined approach called United Cyclic Agree is proposed, which captures the advantages of both theories without generating any new erroneous predictions. In this theory, all of the segments in a probe continue probing in successive cycles until they all have matching goals or there is nowhere left to probe.

A framework for electricity load forecasting based on attention mechanism time series depthwise separable convolutional neural network

Electricity load exhibits daily and weekly cyclical patterns as well as random characteristics. At present, prevailing deep learning models cannot learn electricity load cyclical and stochastic features adequately. This results in insufficient prediction accuracy and the scalability of current methods. To tackle these difficulties, this paper proposes a framework for electrical load prediction based on an Attention Mechanism Time Series Depthwise Separable Convolutional Neural Network (ELPF-ATDSCN). The framework starts by using the Maximum Information Coefficient for exogenous variable selection. It then incorporates a seasonal decomposition algorithm with manual feature engineering to extract the cyclical and stochastic features of the electrical load. Subsequently, the framework employs the ATDSCN to learn the cyclical and stochastic features of the electrical load. In addition, the Bayesian algorithm optimizes model hyperparameters for optimal model performance. Experimental results of point and interval load prediction on datasets from the US and Nordic power markets reveal that the ATDSCN model proposed in this paper enhances load prediction accuracy compared with other models. It can provide more reliable predictions for power system operation and dispatch.

TOP-ReID: Multi-Spectral Object Re-identification with Token Permutation

Multi-spectral object Re-identification (ReID) aims to retrieve specific objects by leveraging complementary information from different image spectra. It delivers great advantages over traditional single-spectral ReID in complex visual environment. However, the significant distribution gap among different image spectra poses great challenges for effective multi-spectral feature representations. In addition, most of current Transformer-based ReID methods only utilize the global feature of class tokens to achieve the holistic retrieval, ignoring the local discriminative ones. To address the above issues, we step further to utilize all the tokens of Transformers and propose a cyclic token permutation framework for multi-spectral object ReID, dubbled TOP-ReID. More specifically, we first deploy a multi-stream deep network based on vision Transformers to preserve distinct information from different image spectra. Then, we propose a Token Permutation Module (TPM) for cyclic multi-spectral feature aggregation. It not only facilitates the spatial feature alignment across different image spectra, but also allows the class token of each spectrum to perceive the local details of other spectra. Meanwhile, we propose a Complementary Reconstruction Module (CRM), which introduces dense token-level reconstruction constraints to reduce the distribution gap across different image spectra. With the above modules, our proposed framework can generate more discriminative multi-spectral features for robust object ReID. Extensive experiments on three ReID benchmarks (i.e., RGBNT201, RGBNT100 and MSVR310) verify the effectiveness of our methods. The code is available at https://github.com/924973292/TOP-ReID.

Dimensions management of traffic big data for short-term traffic prediction on suburban roadways

Since intelligent systems were developed to collect traffic data, this data can be collected at high volume, velocity, and variety, resulting in big traffic data. In previous studies, dealing with the large volume of big traffic data has always been discussed. In this study, big traffic data were used to predict traffic state on a section of suburban road from Karaj to Chalous located in the north of Iran. Due to the many and various extracted features, data dimensions management is necessary. This management was accomplished using principal component analysis to reduce the number of features, genetic algorithms to select features influencing traffic states, and cyclic features to change the nature of features. The data set obtained from each method is used as input to the models. The models used include long short-term memory, support vector machine, and random forest. The results show that using cyclic features can increase traffic state prediction's accuracy than the model, including all the initial features (base model). Long short-term memory model with 71 cyclic features offers the highest accuracy, equivalent to 88.09%. Additionally, this model's reduced number of features led to a shorter modelling execution time, from 456 s (base model) to 201 s.

Enhancing Load Forecasting Accuracy in Smart Grids: A Novel Parallel Multichannel Network Approach Using 1D CNN and Bi‐LSTM Models

Load forecasting plays a pivotal role in the efficient energy management of smart grid. However, the complex, intermittent, and nonlinear smart grids and the complexity of large dataset handling pose difficulty in accurately forecasting loads. The important issue is considering the cyclic features, which have not yet been adequately addressed through the trigonometric transformations. Furthermore, using long short‐term memory (LSTM) or 1D convolution neural network (1D CNN) and existing hybrid models involve stacked CNN‐LSTM architectures, employing 1D convolutions as a preprocessing step to downsample sequences and extract high‐ and low‐level spatial features. However, these models often overlook temporal features, emphasizing higher‐level features processed by the subsequent recurrent neural network layer. Therefore, this study considers a novel approach to independently process features for spatial and temporal characteristics using a parallel multichannel network comprising 1D CNN and bidirectional‐LSTM (Bi‐LSTM) models. The proposed model evaluated the National Transmission and Dispatch Company (NTDC) load dataset, with additional assessment on two datasets, American Electric Power and Commonwealth Edison, to ensure its generalizability. Performance evaluation on the NTDC dataset yields results of 3.4% mean absolute percentage error (MAPE), 513.95 mean absolute error (MAE), and 623.78 root mean square error (RMSE) for day‐ahead forecasting, and 0.56% MAPE, 94.84 MAE, and 115.67 RMSE for hour‐ahead load forecast. The experimental results demonstrate that the proposed model outperforms stacked CNN‐LSTM models, particularly in forecasting hour‐ and day‐ahead loads. Moreover, a comparative analysis with previous studies reveals superior performance in reducing the error gap between predicted and actual values.

Software Implementation and Cyclic Feature Characterization of Nu-FBMC/OQAM Signals

Software Implementation and Cyclic Feature Characterization of Nu-FBMC/OQAM Signals

Accurate Building Energy Management Based on Artificial Intelligence

Abstract Artificial intelligence (AI) plays a pivotal role as a technical facilitator in the domain of building energy management. This paper presents the construction of a sophisticated model for building energy management, leveraging data-driven techniques and machine learning methodologies inherent to AI. The model encompasses three integral modules: characterization of building energy consumption, screening for anomalies in energy consumption, and forecasting of energy usage. Characterization is performed through a data analysis approach. At the same time, anomaly detection utilizes a Gaussian mixture model for data mining and modeling, enabling dynamic and precise identification of states of high energy consumption following clustering analysis. Additionally, the model employs cyclic features and DEEM to accurately predict future energy consumption patterns. The efficacy of this robust energy management model was validated through a case study at Hospital A, located in Changsha, Hunan Province, China. Notably, the total energy consumption at this facility witnessed a modest increase of 1.65% in 2023 compared to the previous year, with a projection of a substantial 24.62% rise in 2024. The surgery building, among various functional units of the hospital, was identified as the most energy-intensive, anticipated to utilize 4.03 million kgce in 2024. Furthermore, the disparity between actual electrical energy and oil consumption in 2023 and those forecasted was contained within 8% and 10%, respectively, underscoring the model’s high predictive accuracy.

Diagnosis of fetal arrhythmia in echocardiography imaging using deep learning with cyclic loss.

Fetal arrhythmias frequently co-occur with congenital heart disease in fetuses. The peaks observed in M-mode fetal echocardiograms serve as pivotal diagnostic markers for fetal arrhythmias. However, speckles, artifacts, and noise pose notable challenges for accurate image analysis. While current deep learning networks mainly overlook cardiac cyclic information, this study concentrated on the integration of such features, leveraging contextual constraints derived from cardiac cyclical features to improve diagnostic accuracy. This study proposed a novel deep learning architecture for diagnosing fetal arrhythmias. The architecture presented a loss function tailored to the cardiac cyclical information and formulated a diagnostic algorithm for classifying fetal arrhythmias. The training and validation processes utilized a dataset comprising 4440 patches gathered from 890 participants. Incorporating cyclic loss significantly enhanced the performance of deep learning networks in predicting peak points for diagnosing fetal arrhythmia, resulting in improvements ranging from 7.11% to 14.81% in F1-score across different network combinations. Particularly noteworthy was the 18.2% improvement in the F1-score for the low-quality group. Additionally, the precision of diagnosing fetal arrhythmia across four categories exhibited improvement, with an average improvement rate of 20.6%. This study introduced a cyclic loss mechanism based on the cardiac cycle information. Comparative evaluations were conducted using baseline methods and state-of-the-art deep learning architectures with the fetal echocardiogram dataset. These evaluations demonstrated the proposed framework's superior accuracy in diagnosing fetal arrhythmias. It is also crucial to note that further external testing is essential to assess the model's generalizability and clinical value.

Generalizable and Discriminative Representations for Adversarially Robust Few-Shot Learning.

Few-shot image classification (FSIC) is beneficial for a variety of real-world scenarios, aiming to construct a recognition system with limited training data. In this article, we extend the original FSIC task by incorporating defense against malicious adversarial examples. This can be an arduous challenge because numerous deep learning-based approaches remain susceptible to adversarial examples, even when trained with ample amounts of data. Previous studies on this problem have predominantly concentrated on the meta-learning framework, which involves sampling numerous few-shot tasks during the training stage. In contrast, we propose a straightforward but effective baseline via learning robust and discriminative representations without tedious meta-task sampling, which can further be generalized to unforeseen adversarial FSIC tasks. Specifically, we introduce an adversarial-aware (AA) mechanism that exploits feature-level distinctions between the legitimate and the adversarial domains to provide supplementary supervision. Moreover, we design a novel adversarial reweighting training strategy to ameliorate the imbalance among adversarial examples. To further enhance the adversarial robustness without compromising discriminative features, we propose the cyclic feature purifier during the postprocessing projection, which can reduce the interference of unforeseen adversarial examples. Furthermore, our method can obtain robust feature embeddings that maintain superior transferability, even when facing cross-domain adversarial examples. Extensive experiments and systematic analyses demonstrate that our method achieves state-of-the-art robustness as well as natural performance among adversarially robust FSIC algorithms on three standard benchmarks by a substantial margin.