Backpropagation-Based Decoding for Multimodal Machine Translation

People are able to describe images using thousands of languages, but languages share only one visual world. The aim of this work is to use the learned intermediate visual representations from a deep convolutional neural network to transfer information across languages for which paired data is not available in any form. Our work proposes using backpropagation-based decoding coupled with transformer-based multilingual-multimodal language models in order to obtain translations between any languages used during training. We particularly show the capabilities of this approach in the translation of German-Japanese and Japanese-German sentence pairs, given a training data of images freely associated with text in English, German, and Japanese but for which no single image contains annotations in both Japanese and German. Moreover, we demonstrate that our approach is also generally useful in the multilingual image captioning task when sentences in a second language are available at test time. The results of our method also compare favorably in the Multi30k dataset against recently proposed methods that are also aiming to leverage images as an intermediate source of translations.

The contribution of reduced COVID-19 test time in controlling the spread of the disease: A simulation-based approach

The novel coronavirus COVID-19 has known a large spread over the globe threatening human health. Recommendations from WHO and specialists insist on testing on a mass scale. However, health systems do not have enough resources. The current process requires the isolation of testees in the hospitals’ isolation rooms for several hours until the test results are revealed, limiting hospitals’ capacities to test large numbers of cases. The aim of this paper was to estimate the impact of reducing the COVID-19 test time on controlling the pandemic spread, through increasing hospitals’ capacities to test on a mass scale. First, a discrete-event simulation was used to model and simulate the COVID-19 testing process in Morocco. Second, a mathematical model was developed to demonstrate the effect of accurate identification of infected cases on controlling the disease’s spread. Simulation results showed that hospitals’ testing capacities could be increased six times if the test duration fell from 10 hours to 10 minutes. The reduction of test time would increase testing capacities, which help to identify all the infected cases. In contrast, the simulation results indicated that if the infected population is not accurately identified and no precautionary measures are taken, the virus will continue to spread until it reaches the total population. Reducing test time is a vital component of the response to the COVID-19 pandemic. It is essential for the effective implementation of policies to contain the virus.

Secure machine learning against adversarial samples at test time

Deep neural networks (DNNs) are widely used to handle many difficult tasks, such as image classification and malware detection, and achieve outstanding performance. However, recent studies on adversarial examples, which have maliciously undetectable perturbations added to their original samples that are indistinguishable by human eyes but mislead the machine learning approaches, show that machine learning models are vulnerable to security attacks. Though various adversarial retraining techniques have been developed in the past few years, none of them is scalable. In this paper, we propose a new iterative adversarial retraining approach to robustify the model and to reduce the effectiveness of adversarial inputs on DNN models. The proposed method retrains the model with both Gaussian noise augmentation and adversarial generation techniques for better generalization. Furthermore, the ensemble model is utilized during the testing phase in order to increase the robust test accuracy. The results from our extensive experiments demonstrate that the proposed approach increases the robustness of the DNN model against various adversarial attacks, specifically, fast gradient sign attack, Carlini and Wagner (C&W) attack, Projected Gradient Descent (PGD) attack, and DeepFool attack. To be precise, the robust classifier obtained by our proposed approach can maintain a performance accuracy of 99% on average on the standard test set. Moreover, we empirically evaluate the runtime of two of the most effective adversarial attacks, i.e., C&W attack and BIM attack, to find that the C&W attack can utilize GPU for faster adversarial example generation than the BIM attack can. For this reason, we further develop a parallel implementation of the proposed approach. This parallel implementation makes the proposed approach scalable for large datasets and complex models.

Doubly Robust Crowdsourcing

Large-scale labeled dataset is the indispensable fuel that ignites the AI revolution as we see today. Most such datasets are constructed using crowdsourcing services such as Amazon Mechanical Turk which provides noisy labels from non-experts at a fair price. The sheer size of such datasets mandates that it is only feasible to collect a few labels per data point. We formulate the problem of test-time label aggregation as a statistical estimation problem of inferring the expected voting score. By imitating workers with supervised learners and using them in a doubly robust estimation framework, we prove that the variance of estimation can be substantially reduced, even if the learner is a poor approximation. Synthetic and real-world experiments show that by combining the doubly robust approach with adaptive worker/item selection rules, we often need much lower label cost to achieve nearly the same accuracy as in the ideal world where all workers label all data points.

Joint Research on Aerodynamic Characteristics and Handling Stability of Racing Car under Different Body Attitudes

With the rapid development of FSAE, the speed of racing cars has increased year by year. As the main research content of racing cars, aerodynamics has received extensive attention from foreign teams. For racing cars, the aerodynamic force on the aerodynamic device ultimately acts on the tires through the transmission of the body and the suspension. When the wheel is subjected to the vertical load generated by the aerodynamic device, the ultimate adhesion capacity of the wheel is improved. Under changing conditions, racing wheels can withstand greater lateral and tangential forces. Therefore, the effects of aerodynamics have a more significant impact on handling stability. The FSAE racing car of Jilin University was taken as the research object, and this paper combines the wind tunnel test, the numerical simulation and the dynamics simulation of the racing system. The closed-loop design process of the aerodynamics of the FSAE racing car was established, and the joint study of aerodynamic characteristics and handling stability of racing car under different body attitudes was realized. Meanwhile, the FSAE car was made the modification of aerodynamic parameter on the basis of handling stability. The results show that, after the modification of the aerodynamic parameters, the critical speed of the car when cornering is increased, the maneuverability of the car is improved, the horoscope test time is reduced by 0.525 s, the downforce of the car is increased by 11.39%, the drag is reduced by 2.85% and the lift-to-drag ratio is increased by 14.70%. Moreover, the pitching moment is reduced by 82.34%, and the aerodynamic characteristics and aerodynamic efficiency of the racing car are obviously improved. On the basis of not changing the shape of the body and the aerodynamic kit, the car is put forward to shorten the running time of the car and improve the comprehensive performance of the car, so as to improve the performance of the car in the race.

Random Forest Model in the Diagnosis of Dementia Patients with Normal Mini-Mental State Examination Scores

Background: Mini-Mental State Examination (MMSE) is the most widely used tool in cognitive screening. Some individuals with normal MMSE scores have extensive cognitive impairment. Systematic neuropsychological assessment should be performed in these patients. This study aimed to optimize the systematic neuropsychological test battery (NTB) by machine learning and develop new classification models for distinguishing mild cognitive impairment (MCI) and dementia among individuals with MMSE ≥ 26. Methods: 375 participants with MMSE ≥ 26 were assigned a diagnosis of cognitively unimpaired (CU) (n = 67), MCI (n = 174), or dementia (n = 134). We compared the performance of five machine learning algorithms, including logistic regression, decision tree, SVM, XGBoost, and random forest (RF), in identifying MCI and dementia. Results: RF performed best in identifying MCI and dementia. Six neuropsychological subtests with high-importance features were selected to form a simplified NTB, and the test time was cut in half. The AUC of the RF model was 0.89 for distinguishing MCI from CU, and 0.84 for distinguishing dementia from nondementia. Conclusions: This simplified cognitive assessment model can be useful for the diagnosis of MCI and dementia in patients with normal MMSE. It not only optimizes the content of cognitive evaluation, but also improves diagnosis and reduces missed diagnosis.

Preliminary evidence of differential expression of myogenic and stress factors in skeletal muscle of older adults with low muscle strength

The age-related loss of muscle strength and mass, or sarcopenia, is a growing concern in the ageing population. Yet, it is not fully understood which molecular mechanisms underlie sarcopenia. Therefore, the present study compared the protein expression profile, such as catabolic, oxidative, stress-related and myogenic pathways, between older adults with preserved (8 ♀ and 5 ♂; 71.5 ±2.6 years) and low muscle strength (6 ♀ and 5 ♂; 78.0±5.0 years). Low muscle strength was defined as chair stand test time >15 seconds and/or handgrip strength <16kg (women) or <27kg (men) according the EWGSOP2 criteria. Catabolic signaling (i.e. FOXO1/3a, MuRF1, MAFbx, LC3b, Atg5, p62) was not differentially expressed between both groups, whereas the mitochondrial marker COX-IV, but not PGC1α and citrate synthase, was lower in the low muscle strength group. Stress factors CHOP and p-ERK1/2 were higher (~1.5-fold) in older adults with low muscle strength. Surprisingly, the inflammatory marker p-p65NF-κB was ~7-fold higher in older adults with preserved muscle strength. Finally, expression of myogenic factors (i.e. Pax7, MyoD, desmin; ~2-fold) was higher in adults with low muscle strength. To conclude, whereas the increased stress factors might reflect the age-related deterioration of tissue homeostasis, e.g. due to misfolded proteins (CHOP), upregulation of myogenic markers in the low strength group might be an attempt to compensate for the gradual loss in muscle quantity and quality. These data might provide valuable insights in the processes that underlie sarcopenia.

FAILURES PREDICTION APPROACH IN AGILE SOFTWARE DEVELOPMENT

Software failure prediction is an important activity during agile software development as it can help managers to identify the failure modules. Thus, it can reduce the test time, cost and assign testing resources efficiently. RapidMiner Studio9.4 has been used to perform all the required steps from preparing the primary data to visualizing the results and evaluating the outputs, as well as verifying and improving them in a unified environment. Two datasets are used in this work, the results for the first one indicate that the percentage of failure to predict the time used in the test is for all 181 rows, for all test times recorded, is 3% for Mean time between failures (MTBF). Whereas, SVM achieved a 97% success in predicting compared to previous work whose results indicated that the use of Administrative Delay Time (ADT) achieved a statistically significant overall success rate of 93.5%. At the same time, the second dataset result indicates that the percentage of failure to predict the time used is 1.5% for MTBF, SVM achieved 98.5% prediction.

Developing Online Learning Assessment Instrument for English Sentence Structure Course during Covid-19 Pandemic

Anchored by the need for constructing an online assessment which is mediated by honesty as the character value for grammar mastery among undergraduate students during Covid-19 pandemic, in the present study we developed a test of English Sentence Structure (ESS) situated in an English department of a public university in Malang, East Java, Indonesia. To enact such a purpose, Design-based Research (DBR) was carried out in the study. Findings from the study showcase that the test was valid and reliable, giving it accessible portion for use in the English department. Aside from that, students also opined positively toward the use of the test in measuring their English grammar mastery. Despite these, we found that students’ score in the tryout phase is low affected by their lack of test preparation, inappropriate situated test time, and ineffective teaching and learning enactment. The paper ends with recommendation for future researchers.