Payments and market infrastructures are the backbone of modern financial systems and play a key role in the economy. One of their main goals is to manage systemic risk, especially in the case of systemically important...
详细信息
Payments and market infrastructures are the backbone of modern financial systems and play a key role in the economy. One of their main goals is to manage systemic risk, especially in the case of systemically important payment systems (SIPS) serving interbank funds transfers. We develop an autoencoder for the Sistema de Pagos Interbancarios (SPI) of Ecuador, which is the largest SIPS, to detect potential anomalies stemming from payment patterns. Our work is similar to Triepels et al. (2018) and Sabetti and Heijmans (2020). We train four different autoencoder models using intraday data structured in three time-intervals for the SPI settlement activity to reconstruct its related payments network. We introduce bank run simulations to feature a baseline scenario and identify relevant autoencoder parametrizations for anomaly detection. The main contribution of our work is training an autoencoder to detect a wide range of anomalies in a payment system, ranging from the unusual behavior of individual banks to systemic changes in the overall structure of the payments network. We also found that these novel techniques are robust enough to support the monitoring of payments' and market infrastructures' functioning, but need to be accompanied by the expert judgement of payments overseers.
We introduce new methods for training an autoencoder (AE) as an unsupervised hyperspectral anomaly detector. We detail a new percentile loss (PL) that reliably constructs an accurate background model while limiting th...
详细信息
ISBN:
(纸本)9781728163741
We introduce new methods for training an autoencoder (AE) as an unsupervised hyperspectral anomaly detector. We detail a new percentile loss (PL) that reliably constructs an accurate background model while limiting the erroneous inclusion of anomalous pixels. We also improve detection performance and reliability by introducing a cumulative detection score that incorporates statistics calculated from the ensemble of AE models generated over the history of the training process. We show improved detection performance on two data sets relative to two baseline algorithms.
We propose a fast few-shot learning framework that uses transfer learning to identify different lung and chest diseases and conditions from chest x-rays. Our model can be trained with as few as five training examples,...
详细信息
ISBN:
(纸本)9781510633964
We propose a fast few-shot learning framework that uses transfer learning to identify different lung and chest diseases and conditions from chest x-rays. Our model can be trained with as few as five training examples, making it potentially applicable for diagnosis of rare diseases. In this work, we divide different chest diseases into two disjoint categories: (i) base classes (with large training set) and (ii) novel classes (with a few training examples per class). Our method consists of two steps, namely feature extraction and classification. For the feature extraction, we employ a deep convolutional neural network, customized for chest x-rays. We train the feature extractor with data only from base classes. So the novel classes are unseen to the feature extractor during training. However, we use the feature extractor for extracting features from the data of novel classes resulting in transfer learning. Our classifier, on the other hand, uses the data only from the novel classes for training. We introduce the idea of autoencoder ensemble to design the classifier. Only a few feature vectors from each of the novel classes are used for training the classifier making it a few-shot learner. Incorporating new novel classes requires training only the classifier which makes the entire process extremely fast. The performance of the classifier is evaluated on the test data from the novel classes. Experiments show 18% improvement in the F-1 score compared to the baseline on identifying the novel diseases from publicly available chest x-ray dataset.
Anomaly detection for textured surface is a key task in product quality control. In recent years, supervised deep learning approaches have begun to be applied in this field, whereas most of the approaches are usually ...
详细信息
ISBN:
(纸本)9781665423144
Anomaly detection for textured surface is a key task in product quality control. In recent years, supervised deep learning approaches have begun to be applied in this field, whereas most of the approaches are usually impracticable in collecting a large scale of defect samples. To this end, this paper proposes an efficient pyramid memory autoencoder. A new approach is developed by reconstructing background texture using autoencoder neural networks with multi-scale spatial pyramid pooling module and affinity memory module. With only positive samples for training phase, the proposed method managed to localize various defects on textured surfaces. Experimental results demonstrated not only its superiority on efficiency and accuracy, but also the great potential in industrial applications.
In this paper, we propose an effective Convolutional autoencoder (AE) model for Sparse Representation (SR) in the Wavelet Domain for Classification (SRWC). The proposed approach involves an autoencoder with a sparse l...
详细信息
ISBN:
(纸本)9781728193205
In this paper, we propose an effective Convolutional autoencoder (AE) model for Sparse Representation (SR) in the Wavelet Domain for Classification (SRWC). The proposed approach involves an autoencoder with a sparse latent layer for learning sparse codes of wavelet features. The estimated sparse codes are used for assigning classes to test samples using a residual-based probabilistic criterion. Intensive experiments carried out on various datasets revealed that the proposed method yields better classification accuracy while exhibiting a significant reduction in the number of network parameters, compared to several recent deep learning-based methods.
Cross-domain image translation attempt to translate images from one domain to another domain, with the content of images preserved. Current approaches treat image's content as the underlying spatial structure, and...
详细信息
ISBN:
(纸本)9781728169262
Cross-domain image translation attempt to translate images from one domain to another domain, with the content of images preserved. Current approaches treat image's content as the underlying spatial structure, and translation only change image's style of color and texture. These methods can generate realistic results, but may not be able to preserve image's fine grained semantic category information and suffer from the lack of diversity in objects' shapes and viewing angles. In this paper, we propose the problem of fine grained category preserving image translation that aims at preserving image's fine grained category information in cross-domain translation. A novel framework called Cross-Domain Adversarial autoencoder (CDAAE) is proposed to solve the problem. CDAAE assumes that cross-domain images have shared content-latent-code space and separate style-latent-code spaces. The content latent code encodes image's basic category information, while the style latent code represents other domain-specific properties, including color, texture, shape, etc. Our experiments evaluate models from aspects of image's quality, diversity as well as category preserving ability, showing CDAAE's advantages over current methods. We also design an algorithm to apply CDAAE to domain adaptation. Experiments on benchmark datasets demonstrate that the proposed method achieves state-of-the-art results.
Luminal-A breast cancer is the most frequently occurring breast cancer subtype. However, it shows high variability in prognosis, and more precise stratification is required for personalized medicine. In this paper, we...
详细信息
ISBN:
(纸本)9781728162157
Luminal-A breast cancer is the most frequently occurring breast cancer subtype. However, it shows high variability in prognosis, and more precise stratification is required for personalized medicine. In this paper, we identify two prognostic subgroups of luminal-A breast cancer. We train a deep autoencoder with gene expression profiles of luminal-A breast cancer, and it automatically generates informative latent features that represent essential properties of gene expressions. We find that two subgroups (BPS-LumA and WPS-LumA) clustered using the latent features are significantly different in prognosis (p-value=1.23e-6;log-rank test). This prognostic difference is validated with other luminal-A breast cancer cohort. The results in our method suggest that the deep autoencoder is able to extract and compress complex properties of gene expressions patterns, and that it is usefully applicable to patient stratification for precision medicine of luminal-A breast cancer.
Dc loads are increasingly used in electric ship power systems supplying a larger amount of power electronic loads. The need for load monitoring and fault detection in dc microgrids has grown but traditional ac methods...
详细信息
ISBN:
(纸本)9781728156354
Dc loads are increasingly used in electric ship power systems supplying a larger amount of power electronic loads. The need for load monitoring and fault detection in dc microgrids has grown but traditional ac methods cannot be easily transferred into dc system. Also, the short-circuit fault is similar to a pulsed load profile which makes it even more difficult in distinguishing shunt faults from normal pulsed power operation. In real-time signal analysis, the neural network receives a majority of attention and parallel computing has allowed for numerical application of neural network research in recent years. The recurrent neural network (RNN) is a specific architecture designed for time-domain signal classification and regeneration. The RNN architecture is different from traditional neural networks in that it keeps a history array buffer of previous signals. The long short-term memory (LSTM) is a type of RNN that addresses gradient propagation issues during network training that are present in vanilla RNN architectures. This paper employs short-time Fourier transform feature extraction and LSTM autoencoder neural networks based classification and fault detection on dc pulsed load monitoring to demonstrate non-intrusive load monitoring applications.
Multi-view feature learning has garnered much attention recently since many real world data are comprised of different representations or views. How to explore the consensus structure and eliminate the inconsistency n...
详细信息
ISBN:
(纸本)9781509066315
Multi-view feature learning has garnered much attention recently since many real world data are comprised of different representations or views. How to explore the consensus structure and eliminate the inconsistency noise in different views remains a challenging problem in multi-view feature learning. In this paper, we propose a multi-way deep autoencoder for multi-view feature learning to explore the deep consensus structure and reconcile the efficiency of encoding process meanwhile. Through a multi-way encoding process, we embed the original data feature views to nonnegative representations of multiple levels which are structured hierarchically. Along the structure of embedded representations, we recover the diversity and important information layer by layer in the decoding process. The experiments on two image datasets show the superior performance of our method.
Visual inspection is a tedious but necessary job in industrial manufacturing to ensure high quality products. Anomaly detection for images is a topic of interest and research, though acquiring anomalous data is diffic...
详细信息
Visual inspection is a tedious but necessary job in industrial manufacturing to ensure high quality products. Anomaly detection for images is a topic of interest and research, though acquiring anomalous data is difficult due to scarcity and labelling. Therefore, unsupervised methods attract attention. This paper proposes an unsupervised method, Defect-Removing autoencoder(DeRA), for anomaly detection using deep learning. The method illustrated to perform exceptionally in various industrial materials and outperformed the state-of-the-art unsupervised anomaly detection methods. Mean AUC is improved on 15 categories of practical applications in MVTec AD from 84.2% to 97.0% compared with previous methods.
暂无评论