In selection processes, decisions follow a sequence of stages. Early stages have more applicants and general information, while later stages have fewer applicants but specific data. This is represented by a dual funne...
详细信息
ISBN:
(纸本)9783030474362;9783030474355
In selection processes, decisions follow a sequence of stages. Early stages have more applicants and general information, while later stages have fewer applicants but specific data. This is represented by a dual funnel structure, in which the sample size decreases from one stage to the other while the information increases. Training classifiers for this case is challenging. In the early stages, the information may not contain distinct patterns to learn, causing underfitting. In later stages, applicants have been filtered out and the small sample can cause overfitting. We redesign the multi-stage problem to address both cases by combining adversarial autoencoders (AAE) and multi-task semi-supervised learning (MTSSL) to train an end-to-end neural network for all stages together. The AAE learns the representation of the data and performs data imputation in missing values. The generated dataset is fed to an MTSSL mechanism that trains all stages together, encouraging related tasks to contribute to each other using a temporal regularization structure. Using real-world data, we show that our approach outperforms other state-ofthe-art methods with a gain of 4x over the standard case and a 12% improvement over the second-best method.
Nowadays, most of the websites like Amazon, YouTube and Netflix use collaborative filtering methods to recommend various types of items to users. There are two principal categories of collaborative filtering;memory-ba...
详细信息
Nowadays, most of the websites like Amazon, YouTube and Netflix use collaborative filtering methods to recommend various types of items to users. There are two principal categories of collaborative filtering;memory-based and model-based. The memory-based methods use the users' similarity measures and have several advantages over the model-based techniques, including being easily explained and easy modeling updates with new ratings and items. However, the memory-based methods' performance reduces when the data is sparse, and unlike the model-based methods, memory-based methods are not scalable. In this paper, we propose a method that exploit the benefits of both similarity-based and model-based approaches. We address both the reliability and the online updating problems based on a novel user-similarity based method. To calculate the new similarity metric we use the predicted user rating vectors in the autoencoder's output and apply mutual information to the predicted vectors in order to find similar users. We depict a similarity graph according to the mutual information rate, which is calculated for each pair of users. We implement the proposed method on the Netflix movie recommendation dataset. According to our experiments, the proposed approach has a significant advantage over the other methods, such as the standard autoencoder, the matrix factorization, and the similarity-based methods. (C) 2021 Elsevier B.V. All rights reserved.
Part-in-whole retrieval (PWR) is an important problem in the field of computer-aided design (CAD) with applications in design reuse, feature recognition and suppression and so on. Initially, we present a non parametri...
详细信息
Part-in-whole retrieval (PWR) is an important problem in the field of computer-aided design (CAD) with applications in design reuse, feature recognition and suppression and so on. Initially, we present a non parametric (and hence threshold independent) algorithm for segmenting CAD models (represented as meshes) which does not require any user intervention. As there is no labelled segmented dataset available for part clustering, we propose the use of autoencoders, one of the approaches used in deep networks along with hierarchical clustering. The features for autoencoder is derived from the Gauss map of the segments. The autoencoder network is then trained and validated using a hierarchical clustering-based approach that generates a dictionary of labels for each segment. PWR is then done by testing a query model with the network that retrieves models having the query as their subset. Comparison of the segmentation algorithm with the state-of-the-art approaches indicate that it performs better or on par. The algorithm was also tested for noisy models. Results of the part clustering and PWR are also presented for models from a CAD dataset along with the discussions. (C) 2019 Elsevier Ltd. All rights reserved.
This paper considers the problem of tracking and predicting the state of a dynamic system with stochastic dynamics and multiple modes of operation. A well-known approach to this problem is the "interacting multip...
详细信息
ISBN:
(纸本)9780738131269
This paper considers the problem of tracking and predicting the state of a dynamic system with stochastic dynamics and multiple modes of operation. A well-known approach to this problem is the "interacting multiple model" (IMM) estimator, which uses knowledge of the different modes of operation to update a bank of Kalman Filters (each optimal for a given mode of operation). The IMM combines estimates according to the posterior probability of the different modes. Despite their popularity, IMMs are known to sometimes be slow to detect mode switching, however, which can result in large state estimation errors. This paper addresses this problem by developing an autoencoder-Interacting Multiple Model (AEIMM) algorithm. The AEIMM effectively embeds an IMM within an autoencoder framework to create a hybrid approach using both deep learning and classical tracking frameworks. The motivation for this approach is that the neural network can perform nonlinear transformations on the measurements to help the IMM more quickly identify mode changes. The effectiveness of the AEIMM is demonstrated in a maneuvering target tracking scenario. Numerical results show that the AEIMM outperforms classical tracking techniques as well as hybrid techniques and a Long Short-Term Memory network in this scenario.
Nowadays, Deep Learning DL becoming more and more interesting in many areas, such as genomics, security, data analysis, image, and video processing. However, DL requires more and more powerful and parallel computing. ...
详细信息
ISBN:
(纸本)9781728175133
Nowadays, Deep Learning DL becoming more and more interesting in many areas, such as genomics, security, data analysis, image, and video processing. However, DL requires more and more powerful and parallel computing. The calculation performed by super-machines equipped with powerful processors, such as the latest GPUs. Despite their power, these computing units consume a lot of energy, which makes their use very difficult in small embedded systems and edge computing. To overcome the problem for which we must keep the maximum performance and satisfy the power constraint, it is necessary to use a heterogeneous strategy. Some solutions are promising when using less energy-consuming electronic circuits, such as FPGAs associated with less expensive topologies such as Stacked Sparse autoencoders. Our target architecture is the Xilinx ZYNQ 7020 SoC, which combines a dual-core ARM processor and an FPGA in the same chip. In the interest of flexibility, we decided to leverage the performance of Xilinx's high-level synthesis tools, evaluate and choose the best solution in terms of size and performance of the data exchange, synchronization and pipeline processing. The results show that our implementation gives high performance at very low energy consumption. Indeed, the evaluation of our accelerator shows that it can classify 1160 MNIST images per second, consuming only 0.443 W;2.4 W for the entire system. More than the low energy consumption and the high performance, the platform used only costs $ 125.
Traffic classification is key for managing both QoS and security in the Internet of Things (IoT). However, new traffic obfuscation techniques have been developed to thwart classification. Traffic mutation is one such ...
详细信息
ISBN:
(数字)9783030457785
ISBN:
(纸本)9783030457785;9783030457778
Traffic classification is key for managing both QoS and security in the Internet of Things (IoT). However, new traffic obfuscation techniques have been developed to thwart classification. Traffic mutation is one such obfuscation technique, that consists of modifying the flow's statistical characteristics to mislead the traffic classifier. In fact, this same technique can also be used to hide normal traffic characteristics for the sake of privacy. However, the concern is its use by attackers to bypass intrusion detection systems by modifying the attack traffic characteristics. In this paper, we propose an unsupervised Deep Learning (DL)-based model to detect mutated traffic. This model is based on generative DL architectures, namely autoencoders (AE) and Generative Adversarial Network (GAN). This model consists of a denoising AE to de-anonymize the mutated traffic and a discriminator to detect it. The implementation results show that the traffic can be denoised when different mutation techniques are applied with a reconstruction error less than 10(-1). In addition, the detection rate of fake traffic reaches 83.7%.
We propose a novel multi-impairment compensation scheme using deep autoencoder for CO-OFDM system. A Q-factor improvement of 10 dB is achieved for QPSK-OFDM signal with 80 km SSMF transmission. Different transmission ...
详细信息
ISBN:
(纸本)9781728154459
We propose a novel multi-impairment compensation scheme using deep autoencoder for CO-OFDM system. A Q-factor improvement of 10 dB is achieved for QPSK-OFDM signal with 80 km SSMF transmission. Different transmission scenarios are also measured to verify the effectiveness.
Noise effects can interfere the face recognition process in outdoor conditions. Therefore, image denoising topic is the classical issue in the field of image processing and computer vision subjects. In this paper, we ...
详细信息
ISBN:
(纸本)9781728145907
Noise effects can interfere the face recognition process in outdoor conditions. Therefore, image denoising topic is the classical issue in the field of image processing and computer vision subjects. In this paper, we show that the solution of denoising process using the autoencoder networks based on the ORL face database. The proposed method can support face recognition systems designed for use in an outdoor environment as the preprocessing stage and it can provide the effective results after training process.
This paper presents an approach for the automated analysis of 3D Computed Tomography (CT) images based on the utilization of descriptors extracted using 3D deep convolutional autoencoder (AEC [8]) networks. Both the c...
详细信息
ISBN:
(数字)9783030582197
ISBN:
(纸本)9783030582180;9783030582197
This paper presents an approach for the automated analysis of 3D Computed Tomography (CT) images based on the utilization of descriptors extracted using 3D deep convolutional autoencoder (AEC [8]) networks. Both the common flow of AEC model application and a set of techniques for overcoming the lack of training samples are presented in this work. The described approach was used for accomplishing the two subtasks of the ImageCLEF 2019: Tuberculosis competition [2,5] and allowed to achieve the 2nd best performance in the TB Severity Scoring subtask and the 6th best performance in the TB CT Report subtask.
With industrial production processes becoming more and more sophisticated, traditional fault diagnosis systems may be insufficient to meet current industrial diagnostic performance requirements. In order to improve fa...
详细信息
ISBN:
(纸本)9781728149646
With industrial production processes becoming more and more sophisticated, traditional fault diagnosis systems may be insufficient to meet current industrial diagnostic performance requirements. In order to improve fault diagnosis performance, this paper proposes an enhanced neural network based fault diagnosis system by integrating Andrews plot and autoencoder. Features are first extracted from on-line measurements by Andrews plot and the high-dimensional features are compressed by autoencoder to an appropriate dimension, which are then fed to a neural network for fault classification. Application to a simulated CSTR process demonstrates that the proposed method can give more reliable and earlier diagnosis than the traditional neural network based fault diagnosis method.
暂无评论