Dual Deep Neural Network Research Articles

Relative likelihood based aggregated dual deep neural network for skin lesion recognition in dermoscopy images

Relative likelihood based aggregated dual deep neural network for skin lesion recognition in dermoscopy images

A hybrid study of a 4-stage compressed solar distiller based on experimental, computational and deep learning methods

In the present study, a hybrid technique is employed to demonstrate the thermal performance of a 4-stage compressed solar distiller with hydrophilic evaporators (CSDHE) regarding the role of natural convection in the air gap domain. Initially, the test is conducted under simulated solar radiation of 2.13 W·m−2 over 5400 s. A thermodynamic model incorporating diffusion is developed for the next step to analyze internal transient temperatures. In the third step, a 3D single-phase natural convection model of the CSDHE is simulated to achieve the research objectives. Based on experimental, thermal modeling, and computational results, a dual deep neural network is developed to predict the thermal behavior of the system and the variation of convection heat transfer coefficient for different air gap thicknesses over 10,000 s. The experimental results revealed a distilled water production rate of 3.806 kg·m−2h−1 with a gained output ratio (GOR) of 112 %. Moreover, the findings demonstrate that conduction accounts for 84.23 % of total heat transfer, while convection accounts for 6.36 % during 5400 s. We also found that, by employing a convolutional neural network (CNN) and directly harvesting data from pre-defined contours, the hybrid training dataset for deep forward neural network (DFNN) is trained 28.35 % faster than conventional methods.

Movie recommendation system via fuzzy decision making based dual deep neural networks

The internet and social networks produce an increasing amount of data. There is a serious necessity for a recommendation system because exploring through the huge collection is time-consuming and difficult. In this study, a multi-modal classifier is introduced which makes use of the output from dual deep neural networks: GRU for text analysis and Faster R-CNN for image analysis. These two networks reduce overall complexity with minimal computational time while retaining accuracy. More precisely, the GRU network is utilized to process movie reviews and the Faster RCNN is used to recognize each frames of the movie trailers. Gated Recurrent Unit (GRU) is a well-known variety of RNN that computes sequential data across recurrent structures. Faster RCNN is an enhanced version of Fast RCNN, it combines with the rectangular region proposals and with the features is extract by the ResNet-101. Initially, the trailer of the movie is manually splitted into frames and these frames are pre-processed using fuzzy elliptical filter for image analysis and the movie reviews are also tokenized for text analysis. The pre-processed text is taken as an input for GRU to classify offensive and non-offensive movies and the pre-processed images are taken as an input for Faster R-CNN to classify violence and non- violence movies based on the extracted features from the movie trailer. Afterwards, the four classified outputs are given as input for fuzzy decision-making unit for recommending best movies based on the Mamdani fuzzy inference system with gauss membership functions. The performance of the dual deep neural networks was evaluated using the specific parameters like specificity, precision, recall, accuracy and F1 score measures. The proposed GRU yields accuracy range of 97.73% for reviews and FRCNN yields the accuracy range of 98.42% for movie trailer.

Sparse I/Q-joint DNN nonlinear equalization based on progressive pruning for a photonics-aided 256-QAM MMW communication system.

An efficient nonlinear equalizer based on the pruning I/Q-joint deep neural network (DNN) is proposed and experimentally demonstrated to mitigate the nonlinearity in a photonics-assisted millimeter-wave (MMW) system with a high-order 256 quadrature-amplitude-modulation (QAM) format. Experimental findings reveal that implementing pruning on the I/Q-joint DNN can compress the computational overhead by 32% while accommodating 256-QAM E-band MMW transmission for a net throughput of 66.67 Gbps with 20.21% less complexity than the traditional Volterra nonlinear equalizer. Compared with the I/Q dual DNN with the same complexity, a 16% pruning ratio improvement is enabled by a robust pruning I/Q-joint DNN that further deciphers the I/Q relationship.

A dual deep neural network for auto-delineation in cervical cancer radiotherapy with clinical validation

BackgroundArtificial intelligence (AI) algorithms are capable of automatically detecting contouring boundaries in medical images. However, the algorithms impact on clinical practice of cervical cancer are unclear. We aimed to develop an AI-assisted system for automatic contouring of the clinical target volume (CTV) and organs-at-risk (OARs) in cervical cancer radiotherapy and conduct clinical-based observations.MethodsWe first retrospectively collected data of 203 patients with cervical cancer from West China Hospital. The proposed method named as SegNet was developed and trained with different data groups. Quantitative metrics and clinical-based grading were used to evaluate differences between several groups of automatic contours. Then, 20 additional cases were conducted to compare the workload and quality of AI-assisted contours with manual delineation from scratch.ResultsFor automatic CTVs, the dice similarity coefficient (DSC) values of the SegNet trained with incorporating multi-group data achieved 0.85 ± 0.02, which was statistically better than the DSC values of SegNet independently trained with the SegNet(A) (0.82 ± 0.04), SegNet(B) (0.82 ± 0.03) or SegNet(C) (0.81 ± 0.04). Moreover, the DSC values of the SegNet and UNet, respectively, 0.85 and 0.82 for the CTV (P < 0.001), 0.93 and 0.92 for the bladder (P = 0.44), 0.84 and 0.81 for the rectum (P = 0.02), 0.89 and 0.84 for the bowel bag (P < 0.001), 0.93 and 0.92 for the right femoral head (P = 0.17), and 0.92 and 0.91 for the left femoral head (P = 0.25). The clinical-based grading also showed that SegNet trained with multi-group data obtained better performance of 352/360 relative to it trained with the SegNet(A) (334/360), SegNet(B) (333/360) or SegNet(C) (320/360). The manual revision time for automatic CTVs (OARs not yet include) was 9.54 ± 2.42 min relative to fully manual delineation with 30.95 ± 15.24 min.ConclusionThe proposed SegNet can improve the performance at automatic delineation for cervical cancer radiotherapy by incorporating multi-group data. It is clinically applicable that the AI-assisted system can shorten manual delineation time at no expense of quality.

Dual Neural Network Architecture for Determining Permeability and Associated Uncertainty

The computation of permeability is vital for reservoir characterization because it is a key parameter in the reservoir models used for estimating and optimizing hydrocarbon production. Permeability is routinely predicted as a correlation from near-wellbore formation properties measured through wireline logs. Several such correlations, namely Schlumberger-Doll Research (SDR) permeability and Timur-Coates permeability models using nuclear magnetic resonance (NMR) measurements, K-lambda using mineralogy, and other variants, have often been used, with moderate success. In addition to permeability, the determination of the uncertainties, both epistemic (model) and aleatoric (data), are important for interpreting variations in the predictions of the reservoir models. In this paper, we demonstrate a novel dual deep neural network framework encompassing a Bayesian neural network (BNN) and an artificial neural network (ANN) for determining accurate permeability values along with associated uncertainties. Deep-learning techniques have been shown to be effective for regression problems but quantifying the uncertainty of their predictions and separating them into the epistemic and aleatoric fractions is still considered challenging. This is especially vital for petrophysical answer products because these algorithms need the ability to flag data from new geological formations that the model was not trained on as “out of distribution” and assign them higher uncertainty. Additionally, the model outputs need sensitivity to heteroscedastic aleatoric noise in the feature space arising due to tool and geological origins. Reducing these uncertainties is key to designing intelligent logging tools and applications, such as automated log interpretation. In this paper, we train a BNN with NMR and mineralogy data to determine permeability with associated epistemic uncertainty, obtained by determining the posterior weight distributions of the network by using variational inference. This provides us the ability to differentiate in- and out-of-distribution predictions, thereby identifying the suitability of the trained models for application in new geological formations. The errors in the prediction of the BNN are fed into a second ANN trained to correlate the predicted uncertainty to the error of the first BNN. Both networks are trained simultaneously and therefore optimized together to estimate permeability and associated uncertainty. The machine-learning permeability model is trained on a “ground-truth” core database and demonstrates considerable improvement over traditional SDR and Timur-Coates permeability models on wells from the Ivar Aasen Field. We also demonstrate the value of information (VOI) of different logging measurements by replacing the logs with their median values from nearby wells and studying the increase in the mean square errors.

A dual deep neural network with phrase structure and attention mechanism for sentiment analysis

Sentiment analysis of short texts is difficult for their simplicity and compactness. This goes a step further when it comes to the Chinese texts. Although deep learning achieved better accuracy in sentiment analysis, there is a lack of explain-ability. Thus, this paper evaluates the effectiveness of techniques for sentiment analysis of Chinese short financial texts with deep learning. For this, we built a Chinese short financial texts corpus (CSFC) and designed an ablation experiment. Beside the CFSC, we used a Chinese review collection and an English short-text repository in the experiment for comparison. There are five techniques involved. They are the Pinyin, the segmentation, the lexical analysis, the phrase structure and the attention mechanism. As results, we found that the phrase structure and the attention mechanism are two of the best. Therefore, the best model in the experiment is called a Phrase Structure and Attention-based Deep network model (PhraSAD). Moreover, to improve the classification accuracy on neutral data, we use a dual classifier strategy for 3-class problems. Experimental results showed that PhraSAD outperformed all other compared models on all experimental datasets.

Dual Deep Neural Network Based Adaptive Filter for Estimating Absolute Longitudinal Speed of Vehicles

This study employs a dual deep neural network (D-DNN) to accurately estimate the absolute longitudinal speed of a vehicle. Accuracy in speed estimation is crucial for vehicle safety, because longitudinal speed is a common parameter employed as a state variable in active safety systems such as anti-lock braking system and traction control system. In this study, DNNs are applied to determine the gain of an adaptive filter to estimate vehicle speed. The used data consists of longitudinal acceleration, wheel speed, filter gain, and estimated vehicle speed. The data generated from Carsim software are collected and preprocessed using a Simulink model. To acquire data with numerous wheel slip patterns, various acceleration and deceleration conditions are applied to four different road conditions. Though, it is challenging to achieve a single DNN model that is optimally cope with the various driving situations. Thus, we adopt two DNN models that were individually trained in low and high acceleration regions. The dual DNN model results in error reductions of 74% and 65%, compared with a single DNN and classical adaptive Kalman filter approaches, respectively.

Dual deep neural network-based classifiers to detect experimental seizures.

Manually reviewing electroencephalograms (EEGs) is labor-intensive and demands automated seizure detection systems. To construct an efficient and robust event detector for experimental seizures from continuous EEG monitoring, we combined spectral analysis and deep neural networks. A deep neural network was trained to discriminate periodograms of 5-sec EEG segments from annotated convulsive seizures and the pre- and post-EEG segments. To use the entire EEG for training, a second network was trained with non-seizure EEGs that were misclassified as seizures by the first network. By sequentially applying the dual deep neural networks and simple pre- and post-processing, our autodetector identified all seizure events in 4,272 h of test EEG traces, with only 6 false positive events, corresponding to 100% sensitivity and 98% positive predictive value. Moreover, with pre-processing to reduce the computational burden, scanning and classifying 8,977 h of training and test EEG datasets took only 2.28 h with a personal computer. These results demonstrate that combining a basic feature extractor with dual deep neural networks and rule-based pre- and post-processing can detect convulsive seizures with great accuracy and low computational burden, highlighting the feasibility of our automated seizure detection algorithm.

Dual Deep Neural Networks Cross-Modal Hashing

Recently, deep hashing methods have attracted much attention in multimedia retrieval task. Some of them can even perform cross-modal retrieval. However, almost all existing deep cross-modal hashing methods are pairwise optimizing methods, which means that they become time-consuming if they are extended to large scale datasets. In this paper, we propose a novel tri-stage deep cross-modal hashing method – Dual Deep Neural Networks Cross-Modal Hashing, i.e., DDCMH, which employs two deep networks to generate hash codes for different modalities. Specifically, in Stage 1, it leverages a single-modal hashing method to generate the initial binary codes of textual modality of training samples; in Stage 2, these binary codes are treated as supervised information to train an image network, which maps visual modality to a binary representation; in Stage 3, the visual modality codes are reconstructed according to a reconstruction procedure, and used as supervised information to train a text network, which generates the binary codes for textual modality. By doing this, DDCMH can make full use of inter-modal information to obtain high quality binary codes, and avoid the problem of pairwise optimization by optimizing different modalities independently. The proposed method can be treated as a framework which can extend any single-modal hashing method to perform cross-modal search task. DDCMH is tested on several benchmark datasets. The results demonstrate that it outperforms both deep and shallow state-of-the-art hashing methods.