Stable, effective and lightweight loop closure detection is an always pursued goal in real-time SLAM systems, that can be ported on embedded processors and deployed on autonomous robotics. Deep learning methods have e...
详细信息
ISBN:
(纸本)9780738133669
Stable, effective and lightweight loop closure detection is an always pursued goal in real-time SLAM systems, that can be ported on embedded processors and deployed on autonomous robotics. Deep learning methods have extended the expressive ability and adaptability of the descriptor, and sequence-based methods can greatly improve the matching accuracy. However, the increased computation complexity and storage bandwidth requirements of matching calculations for high-dimensional descriptor make it infeasible for real-time deployment, especially for robots that navigate in relatively big maps. To address this challenge, we propose a lightweight sequence-based unsupervised loop closure detection scheme. To be specific, Principal Component Analysis (PCA) is applied to squeeze the descriptor dimensions while maintaining sufficient expressive ability. Additionally, with the consideration of the image sequence and combining linear query with fast approximate nearest neighbor search to further reduce the execution time and improve the efficiency of sequence matching. We implement our method on CALC, a state-of-the-art unsupervised solution, and conduct experiments on NVIDIA TX2, results demonstrate that the accuracy has been improved by 5%, while the execution speed is 2x faster.
Cyber-physical critical infrastructures such as power plants are no longer air-gapped. Due to IP-Convergence, the control systems and sensor/actuator communication networks are often directly or indirectly connected t...
详细信息
ISBN:
(数字)9783030780869
ISBN:
(纸本)9783030780869;9783030780852
Cyber-physical critical infrastructures such as power plants are no longer air-gapped. Due to IP-Convergence, the control systems and sensor/actuator communication networks are often directly or indirectly connected to the Internet. While network intrusion detection can provide certain cyber defense capabilities, that is not sufficient due to covert attacks or insider attacks. Therefore, in recent years, a lot of research is being carried out to detect intrusion by observing anomalies in the plants' physical dynamics. In this work, we propose a robust anomaly detection mechanism based on a semi-supervised machine learning technique allowing us near real-time localization of attacks. Deep neural network architecture is used to detect anomaly - based on reconstruction error. We demonstrate our method's efficacy on the SWaT dataset. Our method outperforms other existing anomaly detection techniques with an AUC score of 0.9275.
Identifying cell types is crucial for single-cell RNA sequencing (scRNA-seq) analysis and can be potentially utilized to understand high-level biological processes. Supervised models based on neural networks have rece...
详细信息
ISBN:
(纸本)9780738133669
Identifying cell types is crucial for single-cell RNA sequencing (scRNA-seq) analysis and can be potentially utilized to understand high-level biological processes. Supervised models based on neural networks have recently been successfully applied in the scRNA-seq cell type classification problem and achieved promising results. While most existing works directly use the raw or transformed data, we argue that the original data are too sparse and high-dimensional, and extracting their effective low-dimensional features can better train downstream classifiers, thereby improving the cell type classification performance. In this paper, we propose a novel framework, named Deep Count autoencoder-based Classifier (DCA-CLA), to leverage the discriminative low-dimensional features for classification. Specifically, DCA-CLA first denoises the original count matrix and extracts the data features from the hidden layer using a deep count autoencoder module, then it feeds these bottleneck features into the classifier network to train the learnable parameters and test the performance. Experimental results on eight separate datasets and four pairs of datasets demonstrate that the proposed DCA-CLA framework achieves competitive performance over the state-of-the-art frameworks.
Radio Frequency Fingerprinting (RFF) is one of the promising passive authentication approaches for improving the security of the Internet of Things (IoT). However, with the proliferation of low-power IoT devices, it b...
详细信息
ISBN:
(纸本)9781728133164
Radio Frequency Fingerprinting (RFF) is one of the promising passive authentication approaches for improving the security of the Internet of Things (IoT). However, with the proliferation of low-power IoT devices, it becomes imperative to improve the identification accuracy at low SNR scenarios. To address this problem, this paper proposes a general denoising autoencoder (DAE)-based model for deep learning RFF techniques. Besides, a partially stacking method is designed to appropriately combine the semi-steady and steady-state RFFs of ZigBee devices. The proposed Partially Stacking-based Convolutional DAE (PSC-DAE) aims at reconstructing a high-SNR signal as well as device identification. Experimental results demonstrate that compared to Convolutional Neural Network (CNN), PSCDAE can improve the identification accuracy by 14% to 23.5% at low SNRs (from -10 dB to 5 dB) under Additive White Gaussian Noise (AWGN) corrupted channels. Even at SNR = 10 dB, the identification accuracy is as high as 97.5%.
The performance of speaker recognition systems reduces dramatically in severe conditions in the presence of additive noise and/or reverberation. In some cases, there is only one kind of domain mismatch like additive n...
详细信息
ISBN:
(纸本)9789082797060
The performance of speaker recognition systems reduces dramatically in severe conditions in the presence of additive noise and/or reverberation. In some cases, there is only one kind of domain mismatch like additive noise or reverberation, but in many cases, there are more than one distortion. Finding a solution for domain adaptation in the presence of different distortions is a challenge. In this paper we investigate the situation in which there is none, one or more of the following distortions: early reverberation, full reverberation, additive noise. We propose two configurations to compensate for these distortions. In the first one a specific denoising autoencoder is used for each distortion. In the second configuration, a denoising autoencoder is used to compensate for all of these distortions simultaneously. Our experiments show that, in the co-existence of noise and reverberation, the second configuration gives better results. For example, with the second configuration we obtained 76.6% relative improvement of EER for utterances longer than 12 seconds. For other situations in the presence of only one distortion, the second configuration gives almost the same results achieved by using a specific model for each distortion.
Deep autoregressive models start to become comparable or superior to the conventional systems for automatic speech recognition. However, for the inference computation, they still suffer from inference speed issue due ...
详细信息
ISBN:
(纸本)9781713836902
Deep autoregressive models start to become comparable or superior to the conventional systems for automatic speech recognition. However, for the inference computation, they still suffer from inference speed issue due to their token-by-token decoding characteristic. Non-autoregressive models greatly improve decoding speed by supporting decoding within a constant number of iterations. For example, Align-Refine was proposed to improve the performance of the non-autoregressive system by refining the alignment iteratively. In this work, we propose a new perspective to connect Align-Refine and denoising autoencoder. We introduce a novel noisy distribution to sample the alignment directly instead of obtaining it from the decoder output. The experimental results reveal that the proposed Align-Denoise speeds up both training and inference with performance improvement up to 5% relatively using single-pass decoding.
作者：
Monea, CristianUniv Politehn Bucuresti
Doctoral Sch Elect Telecommun & Informat Technol 1-3 Iuliu Maniu Blvd Bucharest 061071 Romania Mira Technol Grp
Res & Dev Dept 13 Nicolae Grigorescu St Otopeni 075100 Ilfov Romania
Nuclear quadrupole resonance is a highly specific spectroscopy technique for analyzing solid substances with applications ranging from laboratory analysis to security screening screening for prohibited substances. The...
详细信息
Nuclear quadrupole resonance is a highly specific spectroscopy technique for analyzing solid substances with applications ranging from laboratory analysis to security screening screening for prohibited substances. The technique has the drawback of a very low signal-to-noise ratio and multiple signal processing and analysis so-lutions have been proposed for noise rejection and detection. Among these, the deep learning approach using the AlexNet network was recently shown to outperform previous solutions. This paper proposes the enhancement of deep learning detection using transfer learning to extend the applicability of the detection algorithm to other spectrometers and denoising autoencoders to improve its performance at very low signal-to-noise ratios. The transfer learning technique is demonstrated by training the AlexNet network on a simulated data set and transferring the gained knowledge to a real data set. The resulting model achieves a detection accuracy of 98%, close to that obtained by the initial model trained on the real data. Two denoising architectures are proposed, such as deep neural network-based autoencoder and convolutional autoencoder. A comparative evaluation is performed at multiple signal-to-noise ratio conditions in the range [-30, 20] dB, and the convolutional autoencoder is shown to provide the best results, by significantly increasing the detection accuracy by approx. 20% at-30 dB.
This paper proposed a denoising method for modulated signals based on the autoencoder. The auto-encoder is a cascade structure, which is composed of multiple convolution layers and multiple pooling layers. It is mainl...
详细信息
ISBN:
(纸本)9781665449083
This paper proposed a denoising method for modulated signals based on the autoencoder. The auto-encoder is a cascade structure, which is composed of multiple convolution layers and multiple pooling layers. It is mainly divided into a feature encoder and a generation decoder. We use the features of the modulated signal with noise as the input of the auto-encoder and the features of the clean signal as the label. At the same time, back-propagation algorithm and gradient descent method are used to optimize and update the parameters in the auto-encoder model to minimize the reconstruction error, so as to realize the denoising function of the modulated signal. For a variety of modulation types, this method can improve the modulation signal about 3-9 dB in different SNR environment. The denoising model can generate high-level features of different modulation signals without any artificial feature extraction and prior knowledge and has strong feature representation ability. It has the advantages of strong versatility, low complexity, good denoising effect and good stability.
Neural Machine Translation (NMT) tends to perform poorly in low-resource language settings due to the scarcity of parallel data. Instead of relying on inadequate parallel corpora, we can take advantage of monolingual ...
详细信息
ISBN:
(纸本)9781665437530
Neural Machine Translation (NMT) tends to perform poorly in low-resource language settings due to the scarcity of parallel data. Instead of relying on inadequate parallel corpora, we can take advantage of monolingual data available in abundance. Training a denoising self-supervised multilingual sequence-to-sequence model by noising the available large scale monolingual corpora is one way to utilize monolingual data. For a pair of languages for which monolingual data is available in such a pre-trained multilingual denoising model, the model can be fine-tuned with a smaller amount of parallel data from this language pair. This paper presents fine-tuning self-supervised multilingual sequence-to-sequence pre-trained models for extremely low-resource domain-specific NMT settings. We choose one such pre-trained model: mBART. We are the first to implement and demonstrate the viability of non-English centric complete fine-tuning on multilingual sequence-to-sequence pretrained models. We select Sinhala, Tamil and English languages to demonstrate fine-tuning on extremely low-resource settings in the domain of official government documents. Experiments show that our fine-tuned mBART model significantly outperforms state-of-the-art Transformer based NMT models in all pairs in all six bilingual directions, where we report a 4.41 BLEU score increase on Tamil -> Sinhala and a 2.85 BLUE increase on Sinhala -> Tamil translation.
In the industrial applications like fault diagnosis and health management, monitoring data generally reaches sequentially in a streaming form. To recognize fault occurrence in real time without system halt, it is nece...
详细信息
ISBN:
(纸本)9781665440899
In the industrial applications like fault diagnosis and health management, monitoring data generally reaches sequentially in a streaming form. To recognize fault occurrence in real time without system halt, it is necessary to improve the accuracy and stability of anomaly detection with streaming data. To solve this problem, a new online anomaly detection method with streaming data is proposed based on fine-grained feature forecasting. First, to conduct fine-grained decomposition of features, a denoising autoencoder network is run to extract multiple-dimensional deep features of online data in the initial period of normal state. Second, a forecasting model with tensor Tucker decomposition and ARIMA is conducted to predict the fluctuation trend of all feature sequences. Finally, the deviation degree between the prediction values and sequentially-arrived data is calculated, and an alarm threshold is built according to the 95% confidence interval of the maximum deviation. Then the anomalous state data can be detected in real time. This paper adopts the problem of bearing early fault online detection as an example, and run comparative experiments on the IEEE PHM Challenge 2012 bearing dataset. The results show that the proposed method has good detection accuracy and is with no false alarm, while the model training does not rely on any offline data. Then the proposed method is applicable to the problem of online anomaly detection.
暂无评论