Search

The demands of automated shipping address recognition and verification have increased to handle a large number of packages and to save costs associated with misdelivery. A previous study proposed a deep learning system where the shipping address is recognized and verified based on a camera image capturing the shipping address and barcode area. Because the system performance depends on the input image quality, inspection of input image quality is necessary for image preprocessing. In this paper, we propose an input image quality verification method combining global and local features. Object detection and scale-invariant feature transform in different feature spaces are developed to extract global and local features from several independent convolutional neural networks. The conditions of shipping label images are classified by fully connected fusion layers with concatenated global and local features. The experimental results regarding real captured and generated images show that the proposed method achieves better performance than other methods. These results are expected to improve the shipping address recognition and verification system by applying different image preprocessing steps based on the classified conditions., Accepted at ICPR 2020

In the computational prediction of chemical compound properties, molecular descriptors and fingerprints encoded to low dimensional vectors are used. The selection of proper molecular descriptors and fingerprints is both important and challenging as the performance of such models is highly dependent on descriptors. To overcome this challenge, natural language processing models that utilize simplified molecular input line-entry system as input were studied, and several transformer-variant models achieved superior results when compared with conventional methods. In this study, we explored the structural differences of the transformer-variant model and proposed a new self-attention based model. The representation learning performance of the self-attention module was evaluated in a multi-task learning environment using imbalanced chemical datasets. The experiment results showed that our model achieved competitive outcomes on several benchmark datasets. The source code of our experiment is available at https://github.com/arwhirang/sa-mtl and the dataset is available from the same URL., Accepted at ICPR2020

International audience; Early detection of faults in rotating machinery systems is crucial in preventing system failure, increasing safety, and reducing maintenance costs. Current methods of fault detection suffer from the lack of efficient feature extraction method, the need for designating a threshold producing minimal false alarm rates, and the need for expert domain knowledge, which is costly. In this paper, we propose a novel data-driven health division method based on convolutional neural networks using a graphical representation of time series data, called a nested scatter plot. The proposed method trains the model with a small amount of labeled data and does not require a threshold value to predict the health state of rotary machines. Notwithstanding the lack of datasets that show the ground truth of health stages, our experiments with two open datasets of run-to-failure bearing demonstrated that our method is able to detect the early symptoms of bearing wear earlier and more efficiently than other threshold-based health indicator methods.

The data imbalance problem in classification is a frequent but challenging task. In real-world datasets, numerous class distributions are imbalanced and the classification result under such condition reveals extreme bias in the majority data class. Recently, the potential of GAN as a data augmentation method on minority data has been studied. In this paper, we propose a classification enhancement generative adversarial networks (CEGAN) to enhance the quality of generated synthetic minority data and more importantly, to improve the prediction accuracy in data imbalanced condition. In addition, we propose an ambiguity reduction method using the generated synthetic minority data for the case of multiple similar classes that are degenerating the classification accuracy. The proposed method is demonstrated with five benchmark datasets. The results indicate that approximating the real data distribution using CEGAN improves the classification performance significantly in data imbalanced conditions compared with various standard data augmentation methods.

In toxicity evaluation based on the nuclear receptor signalling pathway, in silico prediction tools are used for the detection of the early stages of long-term toxicities, the prioritization of newly synthesized chemicals and the acquisition of the selectivity and sensitivity. Computational prediction model is one of the promising tools for the toxicity screening of the chemical-protein interaction as deep learning has been improved the prediction accuracies. However, the challenge is that data-imbalanced conditions, where the volume of toxic chemical compound dataset is much smaller than the nontoxic dataset, result in low prediction accuracy of the toxic dataset providing valid information to toxicity hazard. In this paper, we have examined the effect of data imbalance in the toxicity assessment data of AR (LBD), ER (LBD), AhR, and PPAR as nuclear receptors, and identified the severe imbalance between the prediction of the toxic and nontoxic datasets. As the acquisition of the balanced selectivity and sensitivity is required for the assessment of toxicity hazards, data resampling methods have been investigated in order to improve the bias problem in binary classification for toxicity hazard profiling of nuclear receptor. The experimental results achieved a sensitivity of 0.714 and a specificity of 0.787, with an overall accuracy of 0.829 and a ROC-AUC of 0.822 by the simple resampling methods.

Shipping labels are widely used in logistics. It is important to ensure the quality of printing label and to verify contents of the shipping label on the package. We developed a verification and recognition method for various types of shipping labels by using deep neural networks. The experimental results showed 96% recognition accuracy in rotation-invariant conditions. Also, we introduce Google Maps API for validating the address which can reduce the cost of returning packages due to the invalid address. To train and evaluate the method, we have generated and collected 25 different types of shipping label dataset. We plan to release the dataset on our website1.

Graphical representation of correlated multi-channel time series signal is studied in this paper. With the proposed multi-layer nested scatter plot (NSP), we can compress multichannel time series signals into static sized data and can obtain advantages inherent in scatter plots.

As data visibility in factories has increased with the deployment of sensors, data-driven maintenance has become popular in industries. Machine learning has been a promising tool for fault detection, but the problem is that the amount of fault data is much less than that of normal data which causes a data imbalance. In this study, we designed a deep neural network for fault detection and diagnosis, and compared the oversampling by a generative adversarial network to standard oversampling techniques. Simulation results indicate that oversampling by the generative adversarial network performs well under the given condition and the deep neural network designed is capable of classifying the faults of an induction motor with high accuracy.

In this study, we developed a novel data-driven fault detection and diagnosis (FDD) method for bearing faults in induction motors where the fault condition data are imbalanced. First, we propose a bearing fault detector based on convolutional neural networks (CNN), in which the vibration signals from a test bench are used as inputs after an image transformation procedure. Experimental results demonstrate that the proposed classifier for FDD performs well (accuracy of 88% to 99%) even when the volume of normal and fault condition data is imbalanced (imbalance ratio varies from 20:1 to 200:1). Additionally, our generative model reduces the level of data imbalance by oversampling. The results improve the accuracy of FDD (by up to 99%) when a severe imbalance ratio (200:1) is assumed.

Applications such as Voice over IP and video streaming require continuous network service, requiring fast failure recovery mechanisms. Proactive failure recovery schemes have been recently proposed to improve network performance during the failure transients. These proactive failure recovery schemes need extra infrastructural support in the form of routing table entries, extra addresses etc. In this paper, we study if the extra infrastructure support can be exploited to build disjoint paths in those frameworks, while keeping the lengths of the recovery paths close to those of the primary paths. Our evaluations show that it is possible to extend the proactive failure recovery schemes to provide support for nearly-disjoint paths which can be employed in multipath routing for load balancing and QoS.

Situational awareness in a disaster is critical to effective response. Disaster responders require timely delivery of high volumes of accurate data to make correct decisions. To meet these needs, we present DistressNet, an ad hoc wireless architecture that supports disaster response with distributed collaborative sensing, topology-aware routing using a multichannel protocol, and accurate resource localization. Sensing suites use collaborative and distributed mechanisms to optimize data collection and minimize total energy use. Message delivery is aided by novel topology management, while congestion is minimized through the use of mediated multichannel radio protocols. Estimation techniques improve localization accuracy in difficult environments.

Reducing energy consumption in networks is receiving considerable research interest. Different approaches have been advocated, from traffic engineering and topology control to hardware-based approaches. These approaches can work at long-time scales of several hours to very small timescales of a few microseconds. In this paper, we focus on finer time granularity energy saving schemes. To increase the usage of active links and to increase the down time of already sleeping links, we use alternative forwarding. To our best knowledge, this is the first attempt to collaborate a routing scheme with hardware scheme to save energy consumption in the wired networks. In our evaluation, alternative forwarding achieves 10% more energy saving of the communication links compared to a hardware-based sleeping scheme.

Several routing protocols have been proposed to take advantage of the dynamic metrics on links such as link delays, queueing lengths, and available link bandwidths. Wardrop routing, QOS routing are few such example routing protocols. Even when these protocols can be shown to offer convergence properties without oscillations, the protocols have not been widely adopted for a number of reasons. The expected cost of keeping the link metrics updated at various nodes in the network being one of them. In this paper, we study the problem of reducing the cost of propagating dynamic link metrics while keeping the quality of paths within a margin of error. We propose a simple technique of threshold-based metric propagation that is shown, through analysis and simulations, to offer bounded guarantees on path quality while significantly reducing the cost of propagating the dynamic link metric information. Our results indicate that threshold based updates can reduce the number of link updates by up to 90-95% in some cases. TRUE pub

Saving energy while providing end-to-end delay guarantees and robust operation have long been regarded as of paramount importance in real-time adhoc and sensor networks. In this paper we explore how rate-adaptation can save energy in ad-hoc and sensor networks that have real-time requirements, and how robustness requirements, achieved by multipath routing, affect the achievable energy savings. We formulate the problem of finding the most energy efficient data rate for each link, propose an adaptive data rate selection algorithm, and demonstrate that our scheme can save up to 15% energy, when compared with state of art, while still meeting the end-to-end delay guarantees.

Applications such as Voice over IP and video delivery require continuous network service, requiring fast failure recovery mechanisms. Proactive Fast failure recovery mechanisms have been recently proposed to improve network performance during the failure transients. The proposed mechanisms need extra infrastructural support in the form of routing table entries, extra addresses etc. In this paper, we study if the extra infrastructure support can be exploited to build disjoint paths in those frameworks, while keeping the recovery path lengths close to the primary paths. Our evaluations show that it is possible to extend the proactive recovery mechanisms to provide support for nearly- disjoint paths.

This paper considers localized packet assignments and distributed packet assignments as joined in the transmission time interval (TTI) for downlink in multi-user OFDM systems. To enhance throughput, we propose an adaptive packet assignment using mode selection between the localized packet assignment and the distributed packet assignment. Moreover, we compare three types of localized packet assignment schemes and show the performance of GPA approaches to optimum performance by improving the spectral efficiency.