Rainfall forecasting is crucial in the field of meteorology and hydrology. However, existing solutions always achieve low prediction accuracy for short-term rainfall forecasting. Numerical forecasting models perform worse in many conditions. Machine learning approaches neglect the influences of physical factors in upstream or downstream regions, which make forecasting accuracy fluctuate in different areas. To improve the overall forecasting accuracy for short-term rainfall, this paper proposes a novel solution called Dynamic Regional Combined short-term rainfall Forecasting approach (DRCF) using Multi-layer Perceptron (MLP). First, Principal Component Analysis (PCA) is used to reduce the dimension of thirteen physical factors, which serves as the input of MLP. Second, a greedy algorithm is applied to determine the structure of MLP. The surrounding sites are perceived based on the forecasting site. Finally, to solve the clutter interference which is caused by the extension of the perception range, DRCF is enhanced with several dynamic strategies. Experiments are conducted on data from 56 real-world meteorology sites in China, and we compare DRCF with atmospheric models and other machine learning approaches. The experimental results show that DRCF outperforms existing approaches in both threat score (TS) and root mean square error (RMSE).

Deep neural networks (DNNs) have been widely and successfully adopted and deployed in various applications of speech recognition. Recently, a few works revealed that these models are vulnerable to backdoor attacks, where the adversaries can implant malicious prediction behaviors into victim models by poisoning their training process. In this paper, we revisit poison-only backdoor attacks against speech recognition. We reveal that existing methods are not stealthy since their trigger patterns are perceptible to humans or machine detection. This limitation is mostly because their trigger patterns are simple noises or separable and distinctive clips. Motivated by these findings, we propose to exploit elements of sound ( e.g ., pitch and timbre) to design more stealthy yet effective poison-only backdoor attacks. Specifically, we insert a short-duration high-pitched signal as the trigger and increase the pitch of remaining audio clips to ‘mask’ it for designing stealthy pitch-based triggers. We manipulate timbre features of victim audio to design the stealthy timbre-based attack and design a voiceprint selection module to facilitate the multi-backdoor attack. Our attacks can generate more ‘natural’ poisoned samples and therefore are more stealthy. Extensive experiments are conducted on benchmark datasets, which verify the effectiveness of our attacks under different settings ( e.g ., all-to-one, all-to-all, clean-label, physical, and multi-backdoor settings) and their stealthiness. Our methods achieve attack success rates of over 95% in most cases and are nearly undetectable. The code for reproducing main experiments are available at https://github.com/HanboCai/BadSpeech_SoE.

During the process of smart city construction, city managers always spend a lot of energy and money for cleaning street garbage due to the random appearances of street garbage. Consequently, visual street cleanliness assessment is particularly important. However, the existing assessment approaches have some clear disadvantages, such as the collection of street garbage information is not automated and street cleanliness information is not real-time. To address these disadvantages, this paper proposes a novel urban street cleanliness assessment approach using mobile edge computing and deep learning. First, the high-resolution cameras installed on vehicles collect the street images. Mobile edge servers are used to store and extract street image information temporarily. Second, these processed street data is transmitted to the cloud data center for analysis through city networks. At the same time, Faster Region-Convolutional Neural Network (Faster R-CNN) is used to identify the street garbage categories and count the number of garbage. Finally, the results are incorporated into the street cleanliness calculation framework to ultimately visualize the street cleanliness levels, which provides convenience for city managers to arrange clean-up personnel effectively. The overall approach is illustrated and visualized using the street images of Jiangning District in Nanjing, China. The practical application shows the feasibility and usability of the approach.

Metamorphic testing has been successfully used in many different fields to solve the test oracle problem. However, how to find a set of appropriate metamorphic relations for metamorphic testing remains a complicated and tedious task. Recently some machine learning approaches have been proposed to predict metamorphic relations. These approaches predicting single label metamorphic relation can alleviate this problem to some extent. However, many applications involve multi-group metamorphic relations, and these approaches are clearly inefficient. To address this problem, in this paper we propose a Multi-Label Metamorphic Relations prediction approach based on an improved radial basis function (RBF) neural network named RBF-MLMR. First, RBF-MLMR uses state-of-the-art soot analysis tool to generate control flow graph and corresponds labels from the source codes of programs. Second, the extracted nodes and the path properties constitute multi-label data sets for the control flow graph. Finally, a multi-label RBF neural network prediction model is established to predict whether the program satisfies multiple metamorphic relations. In order to improve the prediction results, affinity propagation and k-means clustering algorithms are used to optimize the RBF neural network structure of RBF-MLMR. A set of dedicated experiments based on public programs is conducted to validate RBF-MLMR. The experimental results show that RBF-MLMR can achieve accuracy of around 80% for predicting two and three metamorphic relations.

Deep Neural Network (DNN) driven technologies have been extensively employed in various aspects of our life. Nevertheless, the applied DNN always fails to detect erroneous behaviors, which may lead to serious problems. Several approaches have been proposed to enhance adversarial examples for automatically testing deep learning (DL) systems, such as image-based DL systems. However, the approaches contain the following two limitations. First, existing approaches only take into account small perturbations on adversarial examples, they design and generate adversarial examples for a certain particular DNN model. This might hamper the transferability of the examples for other DNN models. Second, they only use shallow features (e.g., pixel-level features) to judge the differences between the generated adversarial examples and the original examples. The deep features, which contain high-level semantic information, such as image object categories and scene semantics, are completely neglected. To address these two problems, we propose CAGFuzz, a Coverage-guided Adversarial Generative Fuzzing testing approach for image-based DL systems. CAGFuzz is able to generate adversarial examples for mainstream DNN models to discover their potential errors. First, we train an Adversarial Example Generator (AEG) based on general datasets. AEG only considers the data characteristics to alleviate the transferability problem. Second, we extract the deep features of the original and adversarial examples, and constrain the adversarial examples by cosine similarity to ensure that the deep features of the adversarial examples remain unchanged. Finally, we use the adversarial examples to retrain the models. Based on several standard datasets, we design a set of dedicated experiments to evaluate CAGFuzz. The experimental results show that CAGFuzz can detect more hidden errors, enhance the accuracy of the target DNN models, and generate adversarial examples with higher transferability