Accelerate adversarial training with loss guided propagation for robust image classification

Adversarial training is effective to train robust image classification models. To improve the robustness, existing approaches often use many propagations to generate adversarial examples, which have high time consumption. In this work, we propose an efficient adversarial training method with loss guided propagation (ATLGP) to accelerate the adversarial training process. ATLGP takes the loss value of generated adversarial examples as guidance to control the number of propagations for each training instance at different training stages, which decreases the computation while keeping the strength of generated adversarial examples. In this way, our method can achieve comparable robustness with less time than traditional training methods. It also has good generalization ability and can be easily combined with other efficient training methods. We conduct comprehensive experiments on CIFAR10 and MNIST, the standard datasets for several benchmarks. The experimental results show that ATLGP reduces 30% to 60% training time compared with other baseline methods while achieving similar robustness against various adversarial attacks. The combination of ATLGP and ATTA (an efficient adversarial training method) achieves superior acceleration potential when robustness meets high requirements. The statistical propagation in different training stages and ablation studies prove the effectiveness of applying loss guided propagation for each training instance. The acceleration technique can more easily extend adversarial training methods to large-scale datasets and more diverse model architectures such as vision transformers.     

基于AHP的高校教师教育技术能力评价指标体系组合权重研究

文章在认真分析《江苏省高校教师教育技术能力指南》基础上,经过归纳建立了高校教师教育技术能力评价指标体系,并应用AHP法(层次分析法)对体系各级指标权重进行了定量研究并给出了一、二级指标的组合权重值。     

粗糙集理论在学生成绩综合评价中的应用

在对学生成绩综合评价中引入粗糙集理论．利用该理论充分挖掘已知数据之间的联系来确定评价指标的权重,避免了传统评价方法所带来的主观性缺陷．计算实例表明了该方法的可行性和有效性．     

位势型算子的任意权弱型(1,1)不等式的完整证明

利用极大算子推广了分数次积分算子的结论,得到了一般位势型算子的两个任意权弱型(1,1)不等式,并给出完整证明。     

基于生态系统理论的农村教师流失成因分析及解决对策

本文基于布朗芬布伦纳的生态系统理论,从微系统、中间系统、外层系统、宏观系统、历时系统五个方面对影响农村教师流失的成因进行了深入探讨,并提出了相应的解决对策。     

Building muscle: nutrition to maximize bulk and strength adaptations to resistance exercise training

Abstract

Several nutritional strategies can optimize muscle bulk and strength adaptations and enhance recovery from heavy training sessions. Adequate energy intake to meet the needs of training and carbohydrate intake sufficient to maintain glycogen stores (>7 g carbohydrate·kg^?1·day^?1 for women; >8 g carbohydrate·kg^?1·day^?1 for men) are important. Dietary protein intake for top sport athletes should include some foods with high biological value, with a maximum requirement of approximately 1.7 g·kg^?1·day^?1 being easily met with an energy sufficient diet. The early provision of carbohydrate (>1 g·kg^?1) and protein (>10 g) early after an exercise session will enhance protein balance and optimize glycogen repletion. Creatine monohydrate supplementation over several days increases body mass through water retention and can increase high-intensity repetitive ergometer performance. Creatine supplementation can enhance total body and lean fat free mass gains during resistance exercise training; however, strength gains do not appear to be enhanced versus an optimal nutritional strategy (immediate post-exercise protein and carbohydrate). Some studies have suggested that β-OH-methyl butyric acid (β-HMB) can enhance gains made through resistance exercise training; however, it has not been compared “head to head” with optimal nutritional practices. Overall, the most effective way to increase strength and bulk is to perform sport-specific resistance exercise training with the provision of adequate energy, carbohydrate, and protein. Creatine monohydrate and β-HMB supplementation may enhance the strength gains made through training by a small margin but the trade-off is likely to be greater bulk, which may be ergolytic for any athlete participating in a weight-supported activity.     

Analysis of the training load during a hypertrophy-type resistance training programme in men and women

Abstract

The purpose of this study was to analyze the specific training load during a resistance training (RT) programme designed to increase muscular hypertrophy in men and women. Thirty-four women (22.7 ± 4.1 years, 58.8 ± 11.9 kg, 162.6 ± 6.2 cm and 22.1 ± 3.6 kg.m^?2) and 30 men (22.7 ± 4.4 years, 68.4 ± 9.0 kg, 174.5 ± 6.6 cm and 22.5 ± 2.4 kg.m^?2) underwent a supervised RT programme that was divided into two phases of 8 weeks each. Training consisted of 10–12 exercises performed with three sets of 8–12 repetitions at repetition maximum resistances performed 3 times per week on nonconsecutive days. There was a significant (P < 0.05) main effect for gender by time interaction for average training load of all the exercises performed in the first 8 weeks of RT with women showing a higher relative increase than men (+43.6% vs. +32.5%, respectively). This result was not observed during the second 8-week phase of the RT programme during which no significant gender by time interaction (P > 0.05) was shown with both genders having a similar relative increase (+28.7% vs. +24.3%, respectively). Women had a higher increase than men in specific average training load of the upper limb exercises during both the first 8 weeks of training (+30.2% vs. +26.6%, respectively) and the second 8 weeks of training (+31.1% vs. +25.3%, respectively). We conclude that the adaptation in specific training load is influenced by gender.     

一种改进的权均值粒子群优化算法

针对粒子群优化算法的早熟收敛和进化后期收敛速度慢等问题,提出了权均值粒子群优化算法。通过在“认知”部分和“社会”部分加入随机权值更新粒子的飞行速度,使粒子能够很快地收敛到全局最优点。典型函数的仿真结果表明,该算法不仅具有较好的全局收敛性能和较快的收敛速度,而且有效地避免了早熟收敛问题。     

论高校图书馆人力资源开发

针对高校图书馆人才流失严重的问题及带来的不良影响进行了反思,并就高校图书馆如何开发人力资源进行了一系列的探讨,提出开发人力资源的六种方法。     

第17号染色体短臂杂合性缺失与肿瘤的研究进展

抑癌基因是人类正常细胞中所具有的一类基因，对细胞的增殖分化有调节作用。染色体体臂上某一区域非随机杂合性缺失导致抑癌基因的丧失，从而促使肿瘤发生。本文总结了第17号染色体短臂(17p)杂合性缺失与肿瘤发生关系的研究进展，该研究不仅为我们定位新的抑癌基因提供了基础，为肿瘤的预防、早期发现和诊断提供了依据，同时也使我们能更好理解肿瘤的分子进展。