Extra Training Research Articles

Robustness meets accuracy in adversarial training for graph autoencoder

Graph autoencoder (GAE) is an effective deep method for graph embedding, while it is vulnerable to the graph adversarial attacks. Adversarial training, which generates adversarial examples in the adversarial scope(neighborhood of natural examples), is effective to improve the robustness of GAE. However, it may lead to degradation of natural accuracy (accuracy on natural examples) due to the extra training examples generated in the adversarial scope (the reasonable scope of adversarial examples). Therefore, considering robustness and natural accuracy is crucial to GAE. In this paper, an improved GAE model is formulated by combining the Structure and Feature encoders, and a novel Adversarial Training strategy (GAE-SFAT) is proposed based on improved GAE. GAE-SFAT has a smaller but more reasonable adversarial scope for adversarial training, which keeps the robustness and reduces the degradation of natural accuracy compared with ordinary adversarial training. In addition, a novel algorithm considering the robustness and accuracy is designed to optimize the GAE-SFAT. We conduct experiments both on the natural graphs as well as perturbed graphs for three datasets. The results show that GAE-SFAT can perform better than state of arts adversarial training model under different kinds of perturbations.

Outcomes of a rural-focussed family practice residency

Purpose: Family medicine physicians play a crucial role in maternity care in community-based settings. However, residency graduates are decreasingly likely to provide delivery services. We explored residency training and graduates' inclusion of obstetrical care in practice by surveying graduates of the Duluth Family Medicine Residency, traditionally producing a high number of graduates serving rural communities and providing obstetrical care.
 Methods: Graduates (N=48) were surveyed about their provision of maternity care and perceptions of how well the residency prepared them to provide such services. Additional factors that impacted decisions to provide obstetric care (e.g., post-residency training, community size, and if there were obstetricians providing delivery services in their community) were evaluated.
 Findings: Nearly half of graduates (46%) were providing maternity services, most of whom provided perinatal clinical care (68% provided delivery care and 4% operative deliveries). Seventy-one percent worked in a community with an available obstetrician. Graduates providing obstetric services were more likely to live in rural communities (66% vs 35%, p<0.01). Thirty percent of graduates received extra obstetrical training in residency. Graduates providing obstetric services were more likely to cite that their decision to practice obstetrics was influenced by the number of deliveries performed, interactions with attending obstetricians, and interactions with prenatal and laboring patients during residency compared to those not providing obstetric services.
 Conclusions: Residency training has an impact on provision of maternity care and potential practice of family physicians in rural communities. Attention to curricular support, particularly exposure to maternity patients, is important for resident comfort in obstetrics.
 

Deep reinforcement learning for large-eddy simulation modeling in wall-bounded turbulence

The development of a reliable subgrid-scale (SGS) model for large-eddy simulation (LES) is of great importance for many scientific and engineering applications. Recently, deep learning approaches have been tested for this purpose using high-fidelity data such as direct numerical simulation (DNS) in a supervised learning process. However, such data are generally not available in practice. Deep reinforcement learning (DRL) using only limited target statistics can be an alternative algorithm in which the training and testing of the model are conducted in the same LES environment. The DRL of turbulence modeling remains challenging owing to its chaotic nature, high dimensionality of the action space, and large computational cost. In this study, we propose a physics-constrained DRL framework that can develop a deep neural network-based SGS model for LES of turbulent channel flow. The DRL models that produce the SGS stress were trained based on the local gradient of the filtered velocities. The developed SGS model automatically satisfies the reflectional invariance and wall boundary conditions without an extra training process so that DRL can quickly find the optimal policy. Furthermore, direct accumulation of reward, spatially and temporally correlated exploration, and the pre-training process are applied for efficient and effective learning. In various environments, our DRL could discover SGS models that produce the viscous and Reynolds stress statistics perfectly consistent with the filtered DNS. By comparing various statistics obtained by the trained models and conventional SGS models, we present a possible interpretation of better performance of the DRL model.

Sperm selection with Annexin-V coated polystrene bead technique (APB-Tech): A novel and reliable method for the microscopic selection of viable and non-apoptotic sperm to be used for intracytoplasmic sperm injection

Infertility is a growing health problem that affects 20% of couples who want to conceive. ‘Intracytoplasmic sperm injection’ (ICSI) is a commonly used assisted reproduction treatment technique that offers the couples to overcome a wide variety of infertility reasons, including female, male and unexplained infertility. A crucial step in ICSI is choosing the correct sperm with the highest potential to form a healthy embryo and thus a healthy offspring. The sperm selection strategies aim to obtain a sperm population with better motility and morphology, which are insufficient to predict a sperm's reproductive potential. The elimination of apoptotic sperms, which is shown to be higher in infertile males and which is shown to affect embryo development and reproductive cohort significantly should also be included in the selection strategies. Among all sperm selection methods, there are only a few which can eliminate apoptotic sperm, but because they need extra types of equipments, a long training period, and high costs, they couldn't find place in the most commonly used techniques in an IVF lab. Selecting the non-apoptotic sperm cells will help us choose a sperm that is more likely to be chosen by the natural selection mechanisms and thus will help to mimic the natural conception more.The study aimed to develop a novel, easy and a harmless individual sperm selection technique to enable choosing non-apoptotic viable sperm cells via light microscopy without any need for extra equipment, education and cost to be used for ''ICSI’. The technique is based on the binding ability of Annexin-V covered polystrene beads to the externalized phosphatidylserine at the outer leaflet of an apoptotic sperm's plasma membrane.After Annexin-V covered polystyrene beads were prepared, beads obtained were attached to BALB-c mice sperm, and the technique is optimized to obtain the most efficient attachment conditions. The results are then compared with the results of four well-known reliable apoptosis detection techniques to test the validity and sperm survival test to test the toxicity of the technique.The method is proven to be '''reliable’ by comparing it with the results of well-known techniques, including TUNEL and SCSA (sperm chromatin structure assay), and '''safe’ by showing its non-toxicity via sperm survival test (SST). In addition, the method enables the selection of sperm cells more closer to naturally-chosen ones from a pool of sperm that should not be allowed to be randomly chosen during microinjection. To date, it was impossible to distinguish a non-apoptotic sperm without harming it or without needing additional equipment other than a routine IVF lab and extra training other than routine andrology work.The technique is named as ‘Annexin-V coated polystyrene bead technique (APB-Tech)'. Based on our results, further studies on APB-Tech should be focused on the possible improvement of ICSI outcomes and, thus, success rates.

Does quality of novice hand-tied square knots improve with repetition during a single training session?

BackgroundKnot tying is a key surgical skill for novices, and repetition over several training sessions improves knot tying. This study examined the effects of repetition within a single training session on quality of knotted loops and compared results of novice trainees and experienced surgeons.MethodsUsing 0.55 mm nylon monofilament fishing line, novices and surgeons each hand-tied 20 knotted loops, using a 2=1=1=1 configuration (surgeon’s throw and three square throws). Loops were mechanically tested with a single load to failure.ResultsAll loops tolerated five newtons (N) preload. More than 70% of novice and surgeon knots failed by slipping or untying, and 8.8% of novice knots and 2.5% of surgeon’s knots were considered dangerous. Surgeons’ loops had less extension at preload, indicating better loop security. However, during single test to failure, there was no difference in mean extension or maximum load between surgeons and novices. There was no significant difference in results of mechanical testing or modes of failure for the first and last ten knotted loops, or the first, second, third, and fourth sets of knotted loops.DiscussionWith appropriate training, novices can construct knots as strong and secure as experienced surgeons. A large percentage of knotted loops are insecure under testing conditions; extra throws may be required when using large diameter monofilament nylon. Novices may require extra training in maintenance of loop security when constructing square knots. Tying more than five or 10 knotted loops within a single training session does not provide added benefits for novices.

Self-Supervised multi-image Super-Resolution Network for Space Image Restoration

Space-based space surveillance (SBSS) has the advantage of being flexible and highly interpretable, allowing it to play a strategic role in monitoring space targets. During actual missions, however, bursts of push-frame imaging methods, dynamic noise during observation, large motions and other factors can result in visible degradation, and it isn’t easy to have the corresponding high-resolution (HR) images for the low-resolution (LR) images acquired at the high-speed intersection, which seriously limits the existing multi-image super-resolution (MISR) methods for space-based applications. To overcome these problems, this paper proposes a self-supervised video super-resolution deep learning method that can be trained end-to-end in space-based observations without LR/HR image pairs. We generate extra training pairs using different sizes of sampling factors for LR videos and train the video hyper-resolution network from coarse to fine using a pyramid format. Extensive experiments on the public satellite dataset BUAA-SID-share1.0 show that our approach outperforms traditional video hyper-resolution methods.

Effect of physical activity combined with extra ciliary-muscle training on visual acuity of children aged 10-11.

This study is intended for exploring the effects of the physical activity combined with extra ciliary-muscle training with different frequencies on children's kinetic visual acuity and uncorrected distance visual acuity, and eventually figuring out the optimal frequency of ciliary-muscle training for each physical education class. To do the present research, A total of 160 students aged 10–11 from a school in Suzhou (a major city located in southeastern Jiangsu Province, East China) were randomly selected and divided into control group (n = 33), 15-frequency group (n = 44), 30-frequency group (n = 40) and 60-frequency group (n = 43), and the latter three experimental groups participated in a specially designed physical activity plan based on the training principles of ciliary muscle, while the control group participated in normal physical activity as usual. The experimental intervention period was 16 weeks, and all students' kinetic visual acuity and uncorrected distance visual acuity were measured before and after the experiment. The result showed that the kinetic visual acuity of the students in the 30 and 60-frequency groups got improved significantly after the experiment (p < 0.05), with the highest improvement occurring in the 30-frequency group, while there was no significant change in the 15-frequency group and the control group; The uncorrected distance visual acuity of the students in the 30 and 60-frequency groups was significantly improved after the experiment (p < 0.05), and the improvement range in these two groups was similar. In contrast, there was no significant change in the 15-frequency group, while the control group showed a significant decrease (p < 0.05). Physical activity combined with extra ciliary-muscle training has a positive effect on improving children's vision; at the same time, ciliary-muscle training with different frequencies bring out different outcomes on children's vision improvement, among which ciliary-muscle training with frequency of 30 in each physical education class is the best choice to enhance children's kinetic visual acuity and uncorrected distance visual acuity.

Relationships between Perfectionism, Extra Training and Academic Performance in Chinese Collegiate Athletes: Mediating Role of Achievement Motivation.

There are limited studies examining the impacts of perfectionism and achievement motivation on collegiate athletes’ extra training and academic achievement in a Chinese context. This study aimed to examine the association of perfectionism (five facets) with extra training and academic performance among Chinese collegiate athletes and identify the mediating role of achievement motivation (two attributes) in the relationship between perfectionism and extra training and academic performance. With a prospective study design, 243 eligible participants completed two-wave surveys from September to December 2021. Measures included demographics, perfectionism (concern over mistake, CM; doubts about action, DA; personal standard, PS; organization; parental expectation, PE), achievement motivation (motive for success, MS; motive for avoiding failure, MF), extra-training (minutes/week), and academic performance (GPA). Results showed that CM, DA, PS, and MS were associated with extra training among Chinese collegiate athletes, while the associations of DA and PS with extra training were mediated by MS. In addition, DA, PS, organization, and MS were associated with participants’ GPA, while MS was a salient mediator for the contributions of DA and PS on participants GPA. Research findings give new insights to the psychological mechanisms of perfectionism and achievement motivation on collegiate athletes’ extra training and academic performance, contributing to future studies in relevant domains.

The role of life skills in developing an authentic leadership attitude in public health students: a multicenter cross-sectional study in Poland

BackgroundDecision-making skills are considered crucial life skills that condition proper social functioning within groups (i.e., support authentic leadership skills and increasing one’s chances of success and wellbeing in life). Nonetheless, the number of scientific papers addressing the role of life skills in developing authentic leadership skills in public health students is limited. The aim of the present study was to develop a theoretical model to determine the role of selected life skills in developing authentic leadership skills in public health students.MethodsThe study was conducted from January 16 through February 28, 2018. In total, 653 students undertaking in-service training in Master’s degree programs qualified for the study, and complete data sets were obtained from 329 students (response rate 50.38%). The data were collected by means of a paper questionnaire. Four research tools were used in the study: The Authentic Leadership Self-Assessment Questionnaire, The Moral Foundations Questionnaire, The General Self-Efficacy Scale, and The Youth Leadership Life Skills Development Scale.ResultsTwo subgroups were identified among the public health students in the study: 1) the extra life skills training group (N = 113) and 2) the no extra life skills training group (N = 216). Both groups of study participants did not differ significantly in terms of age (M (SD): 25.0 (3.89) vs. 25.0 (3.66); t = 0.068, P = 0.946). On the other hand, clear differences were observed in the case of the respondents’ participation in voluntary service. The respondents from the extra life skills training group declared participation in voluntary activities less frequently than the respondents from the second identified group (48.7 vs. 31.9%).ConclusionsA verified theoretical model showed that course aimed at strengthening authentic leadership competences should be modular, should focus on self-improvement and critical reflection, and should be spread over time to enable and encourage each participant to grow and flourish at their own pace.

The Role of Velocity-Based Training (VBT) in Enhancing Athletic Performance in Trained Individuals: A Meta-Analysis of Controlled Trials.

Velocity-based training (VBT) is a rising auto-regulation method that dynamically regulates training loads to promote resistance training. However, the role of VBT in improving various athletic performances is still unclear. Hence, the presented study aimed to examine the role of VBT in improving lower limbs’ maximum strength, strength endurance, jump, and sprint performance among trained individuals. A systematic literature search was performed to identify studies on VBT for lower limb strength training via databases, including PubMed, Web of Science, Embase, EBSCO, Cochrane, CNKI (in Chinese), and Wanfang Database (in Chinese). Controlled trials that deployed VBT only without extra training content were considered. Eventually, nine studies with a total of 253 trained males (at least one year of training experience) were included in the meta-analysis. The pooled results suggest that VBT may effectively enhance lower limbs’ maximum strength (SMD = 0.76; p < 0.001; I2 = 0%), strength endurance (SMD = 1.19; p < 0.001; I2 = 2%), countermovement jump (SMD = 0.53; p < 0.001; I2 = 0%), and sprint ability (SMD of sprint time = −0.40; p < 0.001; I2 = 0%). These findings indicate the positive role of VBT in serving athletic training. Future research is warranted to focus on the effect of velocity loss of VBT on athletic performance.

Nursing students' perceptions of euthanasia legislation: A qualitative study

BackgroundSpain passed a law to legalise euthanasia in March 2021. This law introduces a new right that is incorporated into the portfolio of services provided by the Spanish public health system, meaning that nurses and nursing students will play a key role in its administration. ObjectivesTo determine and understand nursing students' perceptions of the legalisation of euthanasia and the life experiences involved in their positions. DesignThis is a descriptive qualitative study. Settings and participantsThirteen semi-structured, open-ended, focused interviews were conducted with undergraduate nursing students at two faculties in south-eastern Spain. The Atlas ti.9.0 programme was used to analyse their discourses. ResultsThere are students who are in favour of euthanasia and those who are against it. Most students viewed the regulation of conscientious objection as positive and necessary. Students reported that they did not have in-depth knowledge of the law, and those who were familiar with it expressed some misgivings. Students also felt that there was a lack of information and training on the management of euthanasia. ConclusionsThe students described the legalisation of euthanasia as a breakthrough in Spanish public health which will benefit many patients. However, they were aware that they are not capable of facing new legal situation and demanded extra training in ethical conflicts at the end of life to avoid potential psychological sequelae. Nursing faculties should include in their curricula spaces to encourage moral deliberation in complex settings involving the implementation of euthanasia.

Students' readiness to online English language self-learning module and approaches to its integration into the educational process

Introduction. The transition to blended learning in higher education has led to an increase in the share of self-learning and personalized education, which demands taking into account the students' opinions about the variety of teaching methods. The purpose of the article is to analyze the students’ readiness to the integration of the “Online Self-Learning Module”, when learning English and to its implementation approaches. Materials and methods. The main research method is a survey of the 2-3-year students of non-linguistic majors conducted at Togliatti State University (September – November 2021). The questionnaire consisted of 15 questions in three blocks: “Personal goals for self-learning”, “Self-directed learning skills” and “Monitoring of online self-learning process/ Feedback”. Survey results reliability evaluation was carried out using Fisher criterion (φ*-criterion) (p &lt; 0,01). The results. The survey revealed a positive attitude of students to its integration: desire of extra curriculum English training (77%) φ*exp. = 8.068; need for skills of professional texts independent study (65%) φ*exp. = 4.306, personal pace and achievements (67%) φ*exp. = 4.907 Weaknesses, according to the students, include: nonawareness of the teacher-students interaction algorithm (82%) φ*exp. = 9.822; the need for self-assessment and reflection (68%) φ*exp. = 5.204. We determined a new approach to students' self-learning: extra curriculum online task with its modular approach, structurally and thematically correlated with offline learning; the availability of the students’ guidelines and the tasks fulfillment algorithms, the sample for the final answer, and automated evaluation of the assignment assessment. Conclusion. Students identified the “Online Self-Learning Module” for English as a positive trend that will allow them their personalized English learning, to be autonomous in choosing language content, not to depend on the teacher’s direct control, and to increase their self-organization and self-reflection. Integration of “Online Self-Learning Module” involves the educational technologies modernization and serves as a response to the current challenge of the education.

Challenges of English language learning and teaching in Cambodia: A case study of Kith Meng Brasat High School

English has been included in school curricula in many countries, but the challenges of learning and teaching English remain. This study aims to find the challenges of learning and teaching English and solutions to solve the challenges in secondary education in Cambodia. An explanatory sequential mixed-methods design was employed. A total of 250 students (52.40% were females) participated in a survey. The students reported that teachers’ qualities, study programs, learning and teaching materials, classroom activities, class size, and students’ self-learning attitude were the main problems in learning English. A school principal, a deputy principal, and three teachers of English who were interviewed expressed similar concerns regarding the teaching and learning of English. They requested the Ministry of Education, Youth and Sport to provide them with school infrastructure, learning and teaching resources, and extra in-service teacher training workshops. This study recommends that future research with a larger sample size be conducted at other educational institutions across the country.

Requirement-Based Automated Test Case Generation: Systematic Literature Review

There exist multiple techniques of software testing like requirement-based testing (RBT) an approach of software testing from which the tester can generate test cases on the base of requirements without considering the internal system’s structure. In the current area, automation testing is used to minimize time, cost, and human effort. As compared to automated testing, manual testing processes consume more human effort and time. Requirements are documented in natural language so there is no extra training required to understand requirements, RBT is the most used testing technique. Test cases generated with customer requirements are mainly focused on functional test cases. Most approaches focus on real-time embedded systems rather than UML diagrams because non-functional needs are not captured in test cases derived from UML diagrams. Metamodels can be used to extract information from requirements in some cases. Active testing approaches, bounded model checking, activity diagrams, Petri nets round strip strategy, and extended use cases are just a few of the typical ways used to generate test cases. In this article, multiple techniques of automated test case generation have been discussed which are not being addressed in state-of-art literature reviews. Studies included in this systematic literature review (SLR) are built on a set of three research objectives and a variety of high-quality evaluation criteria. Taxonomy has been presented based on test case generation with requirement-based techniques and tools. In the end, gaps and challenges have been discussed to assist researchers to pursue future work.

Ad astra or astray: Exploring linguistic knowledge of multilingual BERT through NLI task

AbstractRecent research has reported that standard fine-tuning approaches can be unstable due to being prone to various sources of randomness, including but not limited to weight initialization, training data order, and hardware. Such brittleness can lead to different evaluation results, prediction confidences, and generalization inconsistency of the same models independently fine-tuned under the same experimental setup. Our paper explores this problem in natural language inference, a common task in benchmarking practices, and extends the ongoing research to the multilingual setting. We propose six novel textual entailment and broad-coverage diagnostic datasets for French, German, and Swedish. Our key findings are that the mBERT model demonstrates fine-tuning instability for categories that involve lexical semantics, logic, and predicate-argument structure and struggles to learn monotonicity, negation, numeracy, and symmetry. We also observe that using extra training data only in English can enhance the generalization performance and fine-tuning stability, which we attribute to the cross-lingual transfer capabilities. However, the ratio of particular features in the additional training data might rather hurt the performance for model instances. We are publicly releasing the datasets, hoping to foster the diagnostic investigation of language models (LMs) in a cross-lingual scenario, particularly in terms of benchmarking, which might promote a more holistic understanding of multilingualism in LMs and cross-lingual knowledge transfer.

Synthetic 18F-FDG PET Image Generation Using a Combination of Biomathematical Modeling and Machine Learning.

Simple SummaryTraining computer-assisted algorithms on medical images, particularly 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) due to its excellent diagnostic accuracy, is difficult, considering small/fragmented samples or privacy concerns. In computer-vision, deep learning-based models, unlike the conventional data augmentation methods, are highly sought after for creating massive medical samples. For this reason, we developed a multi-scale computational framework to generate synthetic 18F-FDG PET images similar to the real ones in different stages of solid tumor growth and angiogenesis. The framework is developed based on the bio-physiological phenomena of FDG radiotracer uptake in solid tumors using a biomathematical model and a generative adversarial network (GAN)-based architecture. The non-invasive augmented 18F-FDG PET images can be used in clinical practice without the need to manage the patient data. In addition, our spatiotemporal mathematical model can calculate the distribution of various radiopharmaceuticals in different tumor-associated vasculatures.No previous works have attempted to combine generative adversarial network (GAN) architectures and the biomathematical modeling of positron emission tomography (PET) radiotracer uptake in tumors to generate extra training samples. Here, we developed a novel computational model to produce synthetic 18F-fluorodeoxyglucose (18F-FDG) PET images of solid tumors in different stages of progression and angiogenesis. First, a comprehensive biomathematical model is employed for creating tumor-induced angiogenesis, intravascular and extravascular fluid flow, as well as modeling of the transport phenomena and reaction processes of 18F-FDG in a tumor microenvironment. Then, a deep convolutional GAN (DCGAN) model is employed for producing synthetic PET images using 170 input images of 18F-FDG uptake in each of 10 different tumor microvascular networks. The interstitial fluid parameters and spatiotemporal distribution of 18F-FDG uptake in tumor and healthy tissues have been compared against previously published numerical and experimental studies, indicating the accuracy of the model. The structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) of the generated PET sample and the experimental one are 0.72 and 28.53, respectively. Our results demonstrate that a combination of biomathematical modeling and GAN-based augmentation models provides a robust framework for the non-invasive and accurate generation of synthetic PET images of solid tumors in different stages.

Dual adversarial learning-based virtual sample generation method for data expansion of soft senors

Many key quality variables are difficult to measure in complex industrial processes for various reasons, such as working conditions or economic costs, leading to inefficient production monitoring. In recent years, soft sensors with outstanding performance in variable estimation have been widely used. However, quality samples collected from industrial sites are often limited, which results in incomplete datasets that cannot meet the training requirements of soft sensors and poor performance in model learning and prediction. In this paper, a new virtual sample generation method DA-GAN based on generative adversarial network (GAN) is proposed to provide extra training samples for soft sensors. Adversarial net-and adversarial sample-based dual adversarial learning is implemented to reduce the adversarial noise in the discriminator gradient, which can improve the convergence speed and learning stability of the generator and obtain virtual samples with higher similarity to the real data. Furthermore, a sample screening method based on asymmetric acceptable domain range expansion is introduced to choose high-quality virtual samples. Experimental results of two industrial case studies show that the virtual samples provided by DA-GAN are closer to real samples than several other widely used generation methods. The performance of the prediction model trained with the dataset added by the virtual samples yielded from DA-GAN can be better improved.

Transformer- and Generative Adversarial Network-Based Inpatient Traditional Chinese Medicine Prescription Recommendation: Development Study.

BackgroundTraditional Chinese medicine (TCM) practitioners usually follow a 4-step evaluation process during patient diagnosis: observation, auscultation, olfaction, inquiry, pulse feeling, and palpation. The information gathered in this process, along with laboratory test results and other measurements such as vital signs, is recorded in the patient’s electronic health record (EHR). In fact, all the information needed to make a treatment plan is contained in the EHR; however, only a seasoned TCM physician could use this information well to make a good treatment plan as the reasoning process is very complicated, and it takes years of practice for a medical graduate to master the reasoning skill. In this digital medicine era, with a deluge of medical data, ever-increasing computing power, and more advanced artificial neural network models, it is not only desirable but also readily possible for a computerized system to mimic the decision-making process of a TCM physician.ObjectiveThis study aims to develop an assistive tool that can predict prescriptions for inpatients in a hospital based on patients’ clinical EHRs.MethodsClinical health records containing medical histories, as well as current symptoms and diagnosis information, were used to train a transformer-based neural network model using the corresponding physician’s prescriptions as the target. This was accomplished by extracting relevant information, such as the patient’s current illness, medicines taken, nursing care given, vital signs, examinations, and laboratory results from the patient’s EHRs. The obtained information was then sorted chronologically to produce a sequence of data for the patient. These time sequence data were then used as input to a modified transformer network, which was chosen as a prescription prediction model. The output of the model was the prescription for the patient. The ultimate goal is for this tool to generate a prescription that matches what an expert TCM physician would prescribe. To alleviate the issue of overfitting, a generative adversarial network was used to augment the training sample data set by generating noise-added samples from the original training samples.ResultsIn total, 21,295 copies of inpatient electronic medical records from Guang’anmen Hospital were used in this study. These records were generated between January 2017 and December 2018, covering 6352 types of medicines. These medicines were sorted into 819 types of first-category medicines based on their class relationships. As shown by the test results, the performance of a fully trained transformer model can have an average precision rate of 80.58% and an average recall rate of 68.49%.ConclusionsAs shown by the preliminary test results, the transformer-based TCM prescription recommendation model outperformed the existing conventional methods. The extra training samples generated by the generative adversarial network help to overcome the overfitting issue, leading to further improved recall and precision rates.

Towards Interpretable Defense Against Adversarial Attacks via Causal Inference

Deep learning-based models are vulnerable to adversarial attacks. Defense against adversarial attacks is essential for sensitive and safety-critical scenarios. However, deep learning methods still lack effective and efficient defense mechanisms against adversarial attacks. Most of the existing methods are just stopgaps for specific adversarial samples. The main obstacle is that how adversarial samples fool the deep learning models is still unclear. The underlying working mechanism of adversarial samples has not been well explored, and it is the bottleneck of adversarial attack defense. In this paper, we build a causal model to interpret the generation and performance of adversarial samples. The self-attention/transformer is adopted as a powerful tool in this causal model. Compared to existing methods, causality enables us to analyze adversarial samples more naturally and intrinsically. Based on this causal model, the working mechanism of adversarial samples is revealed, and instructive analysis is provided. Then, we propose simple and effective adversarial sample detection and recognition methods according to the revealed working mechanism. The causal insights enable us to detect and recognize adversarial samples without any extra model or training. Extensive experiments are conducted to demonstrate the effectiveness of the proposed methods. Our methods outperform the state-of-the-art defense methods under various adversarial attacks.

Effects of Low-Carbohydrate Diet and Exercise Training on Gut Microbiota.

ObjectiveThis study was aimed to evaluate the effects of low-carbohydrate diet (LC) and incorporated high-intensity interval training (HIIT) or moderate-intensity continuous training (MICT) on gut microbiota, and the associations between changes in gut microbiota and cardiometabolic health-related profiles.MethodsFifty overweight/obese Chinese females (age 22.2 ± 3.3 years, body mass index 25.1 ± 3.1 kg/m–2) were randomized to the groups of LC, LC and HIIT (LC-HIIT, 10 repetitions of 6-s sprints and 9-s rest), and LC and MICT group (LC-MICT, cycling at 50–60% V̇O2peak for 30 min). The LC-HIIT and LC-MICT experienced 20 training sessions over 4 weeks.ResultsThe 4-week LC intervention with/without additional training failed to change the Shannon, Chao 1, and Simpson indexes (p > 0.05), LC increased Phascolarctobacterium genus, and LC-HIIT reduced Bifidobacterium genus after intervention (p < 0.05). Groups with extra exercise training increased short-chain fatty acid-producing Blautia genus (p < 0.05) and reduced type 2 diabetes-related genus Alistipes (p < 0.05) compared to LC. Sutterella (r = −0.335) and Enterobacter (r = 0.334) were associated with changes in body composition (p < 0.05). Changes in Ruminococcus, Eubacterium, and Roseburia genera were positively associated with blood pressure (BP) changes (r = 0.392–0.445, p < 0.05), whereas the changes in Bacteroides, Faecalibacterium, and Parabacteroides genera were negatively associated with BP changes (r = −0.567 to −0.362, p < 0.05).ConclusionLC intervention did not change the α-diversity and overall structure of gut microbiota. Combining LC with exercise training may have additional benefits on gut physiology. Specific microbial genera were associated with LC- and exercise-induced regulation of cardiometabolic health.