Search

Explainable Artificial Intelligence (XAI) is targeted at understanding how models perform feature selection and derive their classification decisions. This paper explores post-hoc explanations for deep neural networks in the audio domain. Notably, we present a novel Open Source audio dataset consisting of 30,000 audio samples of English spoken digits which we use for classification tasks on spoken digits and speakers' biological sex. We use the popular XAI technique Layer-wise Relevance Propagation LRP to identify relevant features for two neural network architectures that process either waveform or spectrogram representations of the data. Based on the relevance scores obtained from LRP, hypotheses about the neural networks' feature selection are derived and subsequently tested through systematic manipulations of the input data. Further, we take a step beyond visual explanations and introduce audible heatmaps. We demonstrate the superior interpretability of audible explanations over visual ones in a human user study. • We present a novel audio dataset consisting of 30,000 audio samples of spoken digits. • We use LRP to explain predictions of two different models in the audio domain. • We confirm hypotheses about the neural networks' use of features from explanations. • We present audible explanations and demonstrate their superior interpretability. [ABSTRACT FROM AUTHOR]

For a class of nontriangular nonlinear systems in presence of unknown disturbances, we propose a predictor-based neural dynamic surface control (PNDSC) strategy in this paper. This nontriangular system is transformed via the mean value theorem, and a predictor is then constructed. To avoid an algebraic loop problem, partial state vectors are employed as input signals of neural networks (NNs) for approximating unknown dynamics, and compensation items are designed to compensate for approximation errors from NNs. Different from the traditional NDSC, the PNDSC in this paper utilizes prediction errors to update learning parameters for improving NNs’ learning behaviors with overlarge adaptive gains. On the basis of improved NNs’ approximation behaviors, a predictor-based NNs disturbance observer (PNNDO) is constructed for compensation for external disturbances and approximation errors from NNs. Furthermore, with predictors, a normalization method of weights is developed to reduce the number of online learning parameters. On the basis of the aforementioned result, measurement noises are taken into account in our predictor-based neural control strategy. We employ predictor states, rather than measurement information paralyzed by noises, in design of our control strategy. This reduces high-frequency oscillations in control input. A Lyapunov-based stability analysis shows that all signals are ultimately bounded in the closed-loop system. Finally, the effectiveness of the proposed control strategy is verified by a numerical example and a permanent magnet brushless DC motor system. [ABSTRACT FROM AUTHOR]

An automatic inductor design process is helpful to reduce the design cycle of radio frequency (RF) integrated circuit (IC). This paper proposed an efficient synthesis-analysis machine (SAM) for on-chip inductor synthesis and modeling, as well as an automatic dataset generation (ADG) topology for the generation of artificial neural networks (ANNs) training dataset. The SAM consists of a synthesis ANN, two analysis ANNs and a proposed self-inspection machine (SIM). For a given design request, the SAM synthesizes a layout first, followed by analyzing the performance of the layout automatically, and finally improves the layout confidence through self-inspection. Compared to the modeling functions of analysis ANN, the synthesis ANN behaves as an inverse model, of which the inputs are desired inductor performances while the outputs are geometrical parameters of the layout. However, multi-value problems might occur in obtaining geometries of inductors, suggesting with evidence from mathematic relationships between geometrical and electrical parameters of an inductor. A multi-valued training dataset will mislead the synthesis ANN, resulting in unqualified layouts. To deal with this issue, a solution space contraction technique (SSCM) is also proposed. Furthermore, the SIM suspends most of the failures by comparing the output of analysis ANN back to the input of synthesis ANN. An electromagnetic simulation tool is used for experiments, and the effectiveness of the proposed SAM is proven by 6,500 inductor samples. [ABSTRACT FROM AUTHOR]

The memristor-based neuromorphic computing system (NCS) with emerging storage and computing integration architecture has drawn extensive attention. Because of the unique nonvolatility and programmability, the memristor is an ideal nano-device to realize neural synapses in VLSI circuit implementation of neural networks. However, in the hardware implementation, the performance of the memristive neural network is always affected by quantization error, writing error, and conductance drift, which seriously hinders its applications in practice. In this paper, a novel weight optimization scheme combining quantization and Bayesian inference is proposed to alleviate this problem. Specifically, the weight deviation in the memristive neural network is transformed into the weight uncertainty in the Bayesian neural network, which can make the network insensitive to unexpected weight changes. A quantization regularization term is designed and utilized during the training process of the Bayesian neural network, reducing the quantization error and improving the robustness of the network. Furthermore, a partial training method is raised to extend the applicability of the proposed scheme in large-scale neural networks. Finally, the experiments on a Multilayer Perceptron and LeNet demonstrate that the proposed weight optimization scheme can significantly enhance the robustness of memristive neural networks. [ABSTRACT FROM AUTHOR]

Purpose: This paper aims to compare stochastic gradient method used for neural network training with global optimizer without use of gradient information, in particular differential evolution. Design/methodology/approach: This contribute shows the application of heuristic optimization algorithms to the training phase of artificial neural network whose aim is to predict renewable power production as function of environmental variables such as solar irradiance and temperature. The training problem is cast as the minimization of a cost function whose degrees of freedom are the parameters of the neural network. A differential evolution algorithm is substituted to the more usual gradient-based minimization procedure, and the comparison of their performances is presented. Findings: The two procedures based on stochastic gradient and differential evolution reach the same results being the gradient based moderately quicker in convergence but with a lower value of reliability, as a significant number of runs do not reach convergence. Research limitations/implications: The approach has been applied to two forecasting problems and, even if results are encouraging, the need for extend the approach to other problems is needed. Practical implications: The new approach could open the training of neural network to more stable and general methods, exploiting the potentialities of parallel computing. Originality/value: To the best of the authors' knowledge, the research presented is fully original for the part regarding the neural network training with differential evolution. [ABSTRACT FROM AUTHOR]

Due to the low complexity in arithmetic unit design, stochastic computing (SC) has attracted considerable interest to implement Artificial Neural Networks (ANNs) for resources-limited applications, because ANNs must usually perform a large number of arithmetic operations. To attain a high computation accuracy in an SC-based ANN, extended stochastic logic is utilized together with standard SC units and thus, a stochastic divider is required to perform the conversion between these logic representations. However, the conventional divider incurs in a large computation latency, so limits an SC implementation for ANNs used in applications needing high performance. Therefore, there is a need to design fast stochastic dividers for SC-based ANNs. Recent works (e.g., a binary searching and triple modular redundancy (BS-TMR) based stochastic divider) are targeting a reduction in computation latency, while keeping the same accuracy compared with the traditional design. However, this divider still requires $N$ iterations to deal with $2^{N}$ -bit stochastic sequences, and thus the latency increases in proportion to the sequence length. In this paper, a decimal searching and TMR (DS-TMR) based stochastic divider is initially proposed to further reduce the computation latency; it only requires two iterations to calculate the quotient, so regardless of the sequence length. Moreover, a trade-off design between accuracy and hardware is also presented. An SC-based Multi-Layer Perceptron (MLP) is then considered to show the effectiveness of the proposed dividers over current designs. Results show that when utilizing the proposed dividers, the MLP achieves the lowest computation latency while keeping the same classification accuracy; although incurring in an area increase, the overhead due to the proposed dividers is low over the entire MLP. When using as combined metric for both hardware design and computation complexity the product of the implementation area, latency, power and number of clock cycles, the proposed designs are also shown to be superior to the SC-based MLPs (at the same level of accuracy) employing other dividers found in the technical literature as well as the commonly used 32-bit floating point implementation. [ABSTRACT FROM AUTHOR]

This paper researches the finite-time master–slave synchronization for implicit hybrid neural networks under event-triggered guaranteed cost control and random deception attacks. By developing Lyapunov–Krasovskii functional and exploiting singular value decomposition technique, the finite-time boundedness of synchronization error closed-loop system is achieved under deception attacks and event-triggered scheme, then the finite-time master–slave synchronization of implicit Markovian jump neural networks is realized. The desired gains of state feedback synchronization guaranteed cost controller are designed by solving a set of linear matrix inequalities. The availability of the given approach is confirmed by an artificial neural network and a numerical example. [ABSTRACT FROM AUTHOR]

This paper presents a novel approach to advancing artificial intelligence (AI) through the development of the Complex Recurrent Spectral Network (ℂ -RSN), an innovative variant of the Recurrent Spectral Network (RSN) model. The ℂ -RSN model introduces localized non-linearity, complex fixed eigenvalues, and a distinct separation of memory and input processing functionalities. These features enable the ℂ -RSN to evolve towards a dynamic, oscillating final state that bear some degree of similarity with biological cognition. The model's ability to classify data through a time-dependent function, and the localization of information processing, is demonstrated by using the MNIST dataset. Remarkably, distinct items supplied as a sequential input yield patterns in time which bear the indirect imprint of the insertion order (and of the separation in time between contiguous insertions). • The C-RSN overcomes some limitations in existing neural network models. • C-RSN introduces localized non-linearity, complex eigenvalues and a separated memory. • The network evolves towards a dynamic, oscillating final state. • The model's efficacy is demonstrated through empirical evaluation. • C-RSN is able to process sequential inputs keeping track of the insertion order. [ABSTRACT FROM AUTHOR]

In this paper, we propose a low-power memory-based computing architecture, called selective computing architecture (SCA). It consists of multipliers and an LUT (Look-Up Table)-based component, that is multi-context ternary content-addressable memory (MC-TCAM). Either of them is selected by input-data conditions in neural-networks (NNs). Compared with quantized NNs, a higher accurate multiplication can be performed with low-power consumption in the proposed architecture. If input data stored in the MC-TCAM appears, the corresponding multiplication results for multiple weights are obtained. The MC-TCAM stores only shorter length of input data, resulting in achieving a low-power computing. The performance of the SCA is determined by three physical parameters concerning the configuration of MC-TCAM. The power dissipation of the target NN can be minimized by exploring these parameters in the design space. The hardware based on the proposed architecture is evaluated using TSMC 65 nm CMOS technology and MTJ model. In the case of speech command recognition, the power consumption at the multiplication of the first convolutional layer in a convolutional NN is reduced by 67% compared to the solution relying only on multipliers. [ABSTRACT FROM AUTHOR]

In this paper, we are concerned with the finite-time synchronization of a class of inertial neural networks with time delays. Without applying some finite-time stability theorems, which are widely applied to studying the finite-time synchronization for neural networks, by constructing two Lyapunov functions and using integral inequality method, two sufficient conditions on the finite-time synchronization for a class of inertial neural networks with time delays are derived. Considering that the method and research results of the finite-time synchronization are different from some existing works, this paper extends the works on the finite-time synchronization of neural networks. [ABSTRACT FROM AUTHOR]

In this paper, we propose O4-DNN, a high-performance FPGA-based architecture for convolutional neural network (CNN) accelerators relying on operation packing and out-of-order (OoO) execution for DSP blocks augmented with LUT-based glue logic. The high-level architecture is comprised of a systolic array of processing elements (PEs), supporting output stationary dataflow. In this architecture, the computational unit of each PE is realized by using a DSP block as well as a small number of LUTs. Given the limited number of DSP blocks in FPGAs, the combination (DSP block and some LUTs) provides more computational power obtainable through each DSP block. The proposed computational unit performs eight convolutional operations on five input operands where one of them is an 8-bit weight and the others are four 8-bit input feature (IF) maps. In addition, to improve the energy efficiency of the proposed computational unit, we present an approximate form of the unit suitable for neural network applications. To reduce the memory bandwidth as well as increase the utilization of the computational units, a data reusing technique based on the weight sharing is also presented. To improve the performance of the proposed computational unit further, an addressing approach for computing the partial sums out-of-order is proposed. The efficacy of the architecture is assessed using two FPGA devices executing four state-of-the-art neural networks. Experimental results show that this architecture leads to, on average (up to), $2.5\times $ ($3.44\times$) higher throughput compared to a baseline structure. In addition, on average (maximum of), 12% (40%) energy efficiency improvement is achievable by employing the O4-DNN compared to the baseline structure. [ABSTRACT FROM AUTHOR]

This paper introduces a 3-layer feed-forward neuronet model, trained by novel beetle antennae search weights-and-structure-determination (BASWASD) algorithm. On the one hand, the beetle antennae search (BAS) algorithm is a memetic meta-heuristic optimization algorithm capable of solving combinatorial optimization problems. On the other hand, neuronets trained by a weights-and-structure-determination (WASD) algorithm are known to resolve the shortcomings of traditional back-propagation neuronets, including slow speed of training and local minimum. Combining the BAS and WASD algorithms, a novel BASWASD algorithm is created for training neuronets, and a multi-input BASWASD neuronet (MI-BASWASDN) model is introduced. Using a power sigmoid activation function and while managing the model fitting and validation, the BASWASD algorithm finds the optimal weights and structure of the MI-BASWASDN. Four financial datasets, taken from the European Central Bank publications, validate and demonstrate the MI-BASWASDN model's outstanding learning and predicting performance. Also included is a comparison of the MI-BASWASDN model to three other well-performing neural network models, as well as a MATLAB kit that is publicly available on GitHub to promote and support this research. [ABSTRACT FROM AUTHOR]

This paper proposes a novel end-to-end learning model, called skip-connected covariance (SCCov) network, for remote sensing scene classification (RSSC). The innovative contribution of this paper is to embed two novel modules into the traditional convolutional neural network (CNN) model, i.e., skip connections and covariance pooling. The advantages of newly developed SCCov are twofold. First, by means of the skip connections, the multi-resolution feature maps produced by the CNN are combined together, which provides important benefits to address the presence of large-scale variance in RSSC data sets. Second, by using covariance pooling, we can fully exploit the second-order information contained in such multi-resolution feature maps. This allows the CNN to achieve more representative feature learning when dealing with RSSC problems. Experimental results, conducted using three large-scale benchmark data sets, demonstrate that our newly proposed SCCov network exhibits very competitive or superior classification performance when compared with the current state-of-the-art RSSC techniques, using a much lower amount of parameters. Specifically, our SCCov only needs 10% of the parameters used by its counterparts. [ABSTRACT FROM AUTHOR]

Hardware accelerators for convolution neural networks (CNNs) enable real-time applications of artificial intelligence technology. However, most of the existing designs suffer from low hardware utilization or high area cost due to complex data flow. This paper proposes a hardware efficient vectorwise CNN accelerator that adopts a $3\times 3$ filter optimized systolic array using 1-D broadcast data flow to generate partial sum. This enables easy reconfiguration for different kinds of kernels with interleaved input or elementwise input data flow. This simple and regular data flow results in low area cost while attains high hardware utilization. The presented design achieves 99%, 97%, 93.7%, and 94% hardware utilization for VGG-16, ResNet-34, GoogLeNet, and Mobilenet, respectively. Hardware implementation with TSMC 40nm technology takes 266.9K NAND gate count and 191KB SRAM to support 168GOPS throughput while consumes only 154.98mW when running at 500MHz operating frequency, which has superior area and power efficiency than other designs. [ABSTRACT FROM AUTHOR]

Transcendental functions are commonly used in many fields such as nonlinear functions of artificial neural networks (ANNs). Due to nonlinearity of these functions, hardware implementation of the functions faces many challenges. Some methods cannot control the approximation accuracy beforehand, or they cannot approximate target functions accurately and efficiently. Other methods can only approximate a particular function without generality. A universal piecewise linear (PWL) approximation method is proposed to solve the problems existing in the above methods. In this paper, a general piecewise linear (PWL) approximation method with controllable maximum absolute error for transcend-ental functions is proposed. The method has the self-adaptive capability to choose the smallest number of segments under the constraint of a controllable maximum absolute error. Therefore, it requires fewer segments and incurs less hardware overhead. Moreover, it can approximate any transcendental functions and does not rely on any properties of the target function. Comparing our work with other methods used in three transcendental functions, including the sigmoid, hyperbolic tangent, and logarithmic function, our method improves the approximation accuracy by 4.5 times, 2.1 times, and 1.3 times, respectively; in addition, our method is more efficient and it reduces the circuit area by 49.9%, 22.4%, and 28.5%, respectively. [ABSTRACT FROM AUTHOR]

• Fixed-time synchronization of a class of memristor-based neural networks were investigated. • Two control schemes are proposed to achieve fixed-time synchronization. • The impulsive effects are overcomed. • Finally, an example and its simulations are given to demonstrate the feasibility of the obtained results. This paper concerns the fixed-time synchronization of a class of memristor-based neural networks with impulsive effects. Since impulses may destroy the synchronization, it is indispensable to design suitable controllers to control its unpredictable effects. In the present paper, two control schemes are proposed to achieve fixed-time synchronization of the impulsive memristor-based neural networks. Finally, two examples and their numerical simulations are presented to show the effectiveness of the theoretical results. [ABSTRACT FROM AUTHOR]

In this paper, a neural network approach is suggested to the product design analysis. Namely, fractional-order neural network models are proposed as more flexible mechanism to study product form design. Since control and stability methods are fundamental in the construction and practical significance of a neural network model, appropriate controllers are designed and practical stability criteria are proposed for the fractional-order model under consideration. The stability and control analysis are based on the Lyapunov function method. Examples are elaborated to demonstrate the established results. The proposed modeling approach and the stability results are also applicable to numerous industrial design tasks. • Fractional-order neural network modeling approach is introduced to the product form design. • Appropriate controllers are utilized. • The practical stability notion is adopted to the introduced model. • Practical stability criteria are established using the Lyapunov function technique. • Examples and discussion are also offered to verify and justify the proposed results. [ABSTRACT FROM AUTHOR]

Purpose: The purpose of this study aims to focus on the detection and identification of the broken rotor bars (BRBs) of a squirrel cage induction motor (SCIM). The presented diagnosis technique is based on artificial neural networks (NNs) that use as inputs the results of the spectral analysis using the fast Fourier transform (FFT) of the reduced Park's vector modulus (RPVM), along with the load values in which the motor operates. Design/methodology/approach: First, this paper presents a comparative study between FFT applied on Hilbert modulus, Park's vector modulus and RPVM to extract feature frequencies of BRB faults. Moreover, the extracted features of FFT applied to RPVM and the load values were selected as NNs' inputs for the detection of the number of BRBs. Findings: The obtained simulation results using MATLAB (Matrix Laboratory) environment show the effectiveness and accuracy of the proposed NNs based approach. Originality/value: The current paper presents a novel diagnostic method for BRBs' fault detection in SCIM, based on the combination between the signal processing analysis (FFT of RPVM) and artificial intelligence (NNs). [ABSTRACT FROM AUTHOR]

In this paper, the receding horizon near-optimal tracking control problem about a class of continuous-time nonlinear systems with fully unknown dynamics is considered. The main challenges of this problem lie in two aspects: 1) most existing systems only restrict their considerations to the state feedback part while the input channel parameters are assumed to be known. This paper considers fully unknown system dynamics in both the state feedback channel and the input channel and 2) the optimal control of nonlinear systems requires the solution of nonlinear Hamilton–Jacobi–Bellman equations. Up to date, there are no systematic approaches in the existing literature to solve it accurately. A novel model-free adaptive near-optimal control method is proposed to solve this problem via utilizing the Taylor expansion-based problem relaxation, the universal approximation property of sigmoid neural networks, and the concept of sliding mode control. By making approximation for the performance index, it is first relaxed to a quadratic program, and then, a linear algebraic equation with unknown terms. An auxiliary system is designed to reconstruct the input-to-output property of the control systems with unknown dynamics, so as to tackle the difficulty caused by the unknown terms. Then, by considering the property of the sliding-mode surface, an explicit adaptive near-optimal control law is derived from the linear algebraic equation. Theoretical analysis shows that the auxiliary system is convergent, the resultant closed-loop system is asymptotically stable, and the performance index asymptomatically converges to optimal. An illustrative example and experimental results are presented, which substantiate the efficacy of the proposed method and verify the theoretical results. [ABSTRACT FROM AUTHOR]

In this paper, we deal with the learning problem when using an adaptive filtering method. For the learning system in filtering, the knowledge is obtained and updated based on the newly acquired information that is extracted and learned from the sequential samples over time. Effective measurement on the informativeness of a sample and reasonable subsequent treatment on the sample will improve the learning performance. This paper proposes a sequential outlier criterion for sparsification of online adaptive filtering. The method is proposed to achieve effective informativeness measurement of online filtering to obtain a more accurate and more compact network in the learning process. In the proposed method, the measurement on the samples’ informativeness is established based on the historical sequentially adjacent samples, and then the informative-measured samples are treated individually by the learning system based on whether the sample is informative, redundant, or abnormal. With our method, a more sensible learning process can be achieved with valid knowledge extracted, and the optimal network in the learning system can be obtained. Simulations based on static function estimation, Mackey-Glass time series prediction, and Lorenz chaotic time series prediction demonstrate that the proposed method can provide more effective classification on samples and more accurate networks in online adaptive filtering. [ABSTRACT FROM AUTHOR]

This paper proposes a novel robust zeroing neural-dynamics (RZND) approach as well as its associated model for solving the inverse kinematics problem of mobile robot manipulators. Unlike existing works based on the assumption that neural network models are free of external disturbances, four common forms of time-varying disturbances suppressed by the proposed RZND model are investigated in this paper. In addition, theoretical analyses on the antidisturbance performance are presented in detail to prove the effectiveness and robustness of the proposed RZND model with time-varying disturbances suppressed for solving the inverse kinematics problem of mobile robot manipulators. That is, the RZND model converges toward the exact solution of the inverse kinematics problem of mobile robot manipulators with bounded or zero-oriented steady-state position error. Moreover, simulation studies and comprehensive comparisons with existing neural network models, e.g., the conventional Zhang neural network model and the gradient-based recurrent neural network model, together with extensive tests with four common forms of time-varying disturbances substantiate the efficacy, robustness, and superiority of the proposed RZND approach as well as its time-varying disturbances suppression model for solving the inverse kinematics problem of mobile robot manipulators. [ABSTRACT FROM AUTHOR]

This paper presents a novel method for learning the cyclic contents of stochastic time series: the deep time-growing neural network (DTGNN). The DTGNN combines supervised and unsupervised methods in different levels of learning for an enhanced performance. It is employed by a multiscale learning structure to classify cyclic time series (CTS), in which the dynamic contents of the time series are preserved in an efficient manner. This paper suggests a systematic procedure for finding the design parameter of the classification method for a one-versus-multiple class application. A novel validation method is also suggested for evaluating the structural risk, both in a quantitative and a qualitative manner. The effect of the DTGNN on the performance of the classifier is statistically validated through the repeated random subsampling using different sets of CTS, from different medical applications. The validation involves four medical databases, comprised of 108 recordings of the electroencephalogram signal, 90 recordings of the electromyogram signal, 130 recordings of the heart sound signal, and 50 recordings of the respiratory sound signal. Results of the statistical validations show that the DTGNN significantly improves the performance of the classification and also exhibits an optimal structural risk. [ABSTRACT FROM AUTHOR]

In this paper, a differential neural network (DNN) implemented as a robust observer estimates the dynamics of perturbed uncertain nonlinear systems affected by exogenous unknown inputs. In the first stage, the identification error converges into a neighborhood around the origin. Then, the second-order sliding mode supertwisting algorithm implemented as a robust exact differentiator reconstructed the unknown inputs. The approach proposed in this paper can be applied in the case of full access to the state vector (identification problem) and in the case of partial access to the state vector (estimation problem). In the second case, the nonlinear system under study must have well-defined full relative degree with respect to the unknown input. Numerical examples showed the effectiveness of the proposed algorithm. The first example tested the DNN working as an identifier into a mathematical model describing the dynamics of a spatial minisatellite. The second example (with a DNN implemented as an observer) tested the methodology of this paper over a single link flexible robot manipulator represented in a canonical (Brunovsky) form. In both examples, the mathematical models served as data generators in the testing of the neural networks. Even when not exact mathematical description of both models was used in the input estimation, the accuracy obtained with the DNN is comparable with the case of applying a high-order differentiator with complete knowledge of the plant. [ABSTRACT FROM AUTHOR]

Quantile regression is widely applied in various fields such as economy, energy, meteorological prediction research in recent years since it does not require distribution assumptions, has relatively loose conditions, and can effectively estimate the uncertainty of time series forecasting. In this paper, a monotone quantile regression neural network (MQRNN) framework is constructed for time series quantile forecasting. The proposed approach takes the monotonicity of quantile into consideration and handles the quantile crossing problem by adding the quantile information into the input structure and using the gradient based point-wise loss function. Aiming at the complex characteristics of time series, such as time-varying and asymmetric heavy-tailed features, a new quantile function is utilized to describe the complete conditional distribution information of data. Under this model framework, non-crossing multiple quantiles can be predicted simultaneously. The proposed approach is implemented based on artificial neural networks, and the constructed model is applied to actual data in different fields. The experimental results demonstrate that the proposed method combined with long short-term memory (LSTM) can provide accurate and reliable multi-quantile prediction, and alleviate the problem of quantile crossing. [ABSTRACT FROM AUTHOR]

In this paper, a new global exponential stability criterion is obtained for a general multidimensional delay difference equation using induction arguments. In the cases that the difference equation is periodic, we prove the existence of a periodic solution by constructing a type of Poincaré map. The results are used to obtain stability criteria for a general discrete-time neural network model with a delay in the leakage terms. As particular cases, we obtain new stability criteria for neural network models recently studied in the literature, in particular for low-order and high-order Hopfield and Bidirectional Associative Memory (BAM). • Global exponential stability criterion for a general delay difference system. • New method to prove global exponential stability of a delay difference system. • Existence of a periodic solution of a general delay difference system. • Exponential stability criteria for neural networks with delay in leakage terms. • Existence of periodic solution for neural networks with delay in the leakage terms. [ABSTRACT FROM AUTHOR]

Hashing is emerging as a powerful tool for building highly efficient indices in large-scale search systems. In this paper, we study spectral hashing (SH), which is a classical method of unsupervised hashing. In general, SH solves for the hash codes by minimizing an objective function that tries to preserve the similarity structure of the data given. Although computationally simple, very often SH performs unsatisfactorily and lags distinctly behind the state-of-the-art methods. We observe that the inferior performance of SH is mainly due to its imperfect formulation; that is, the optimization of the minimization problem in SH actually cannot ensure that the similarity structure of the high-dimensional data is really preserved in the low-dimensional hash code space. In this paper, we, therefore, introduce reversed SH (ReSH), which is SH with its input and output interchanged. Unlike SH, which estimates the similarity structure from the given high-dimensional data, our ReSH defines the similarities between data points according to the unknown low-dimensional hash codes. Equipped with such a reversal mechanism, ReSH can seamlessly overcome the drawback of SH. More precisely, the minimization problem in our ReSH can be optimized if and only if similar data points are mapped to adjacent hash codes, and mostly important, dissimilar data points are considerably separated from each other in the code space. Finally, we solve the minimization problem in ReSH by multilayer neural networks and obtain state-of-the-art retrieval results on three benchmark data sets. [ABSTRACT FROM AUTHOR]

Hardware design of deep convolutional neural networks (CNNs) faces challenges of high computational complexity and data bandwidth as well as huge divergence in different CNN network layers, in which the throughput of the convolutional layer would be bounded by available hardware resource, and throughput of the fully connected layer would be bounded by available data bandwidth. Thus, a highly flexible and efficient design is desired to meet these needs. This paper presents an end-to-end CNN accelerator that maximizes hardware utilization with run-time configurations of different kernel sizes. It also minimizes data bandwidth with the output first strategy to improve the data reuse of the convolutional layers by up to $300\times \sim 600\times $ compared with the non-reused case. The whole CNN implementation of the target network is generated optimally for both hardware and data efficiency under design resource constraints, which can be run-time reconfigured by the layer optimized parameters to achieve real-time and end-to-end CNN acceleration. An implementation example for AlexNet consumes a 1.783 M gate count for 216 MACs and a 142.64 kb internal buffer with TSMC 40 nm process, and achieves 99.7 and 61.6 f/s under 454 MHz clock frequency for the convolutional layers and the whole AlexNet, respectively. [ABSTRACT FROM PUBLISHER]

In the past few years, the demand for real-time hardware implementations of deep neural networks (DNNs), especially convolutional neural networks (CNNs), has dramatically increased, thanks to their excellent performance on a wide range of recognition and classification tasks. When considering real-time action recognition and video/image classification systems, latency is of paramount importance. Therefore, applications strive to maximize the accuracy while keeping the latency under a given application-specific maximum: in most cases, this threshold cannot exceed a few hundred milliseconds. Until now, the research on DNNs has mainly focused on achieving a better classification or recognition accuracy, whereas very few works in literature take in account the computational complexity of the model. In this paper, we propose an efficient computational method, which is inspired by a computational core of fully connected neural networks, to process convolutional layers of state-of-the-art deep CNNs within strict latency requirements. To this end, we implemented our method customized for VGG and VGG-based networks which have shown state-of-the-art performance on different classification/recognition data sets. The implementation results in 65-nm CMOS technology show that the proposed accelerator can process convolutional layers of VGGNet up to 9.5 times faster than state-of-the-art accelerators reported to-date while occupying 3.5 mm2. [ABSTRACT FROM AUTHOR]

This paper is concerned with the stability analysis of neural networks with a time-varying delay. To assess system stability accurately, the conservatism reduction of stability criteria has attracted many efforts, among which estimating integral terms as exact as possible is a key issue. At first, this paper develops a new relaxed integral inequality to reduce the estimation gap of popular Wirtinger-based inequality (WBI). Then, for showing the advantages of the proposed inequality over several existing inequalities that also improve the WBI, four stability criteria are derived through different inequalities and the same Lyapunov–Krasovskii functional (LKF), and the conservatism comparison of them is analyzed theoretically. Moreover, an improved criterion is established by combining the proposed inequality and an augmented LKF with delay-product-type terms. Finally, several numerical examples are used to demonstrate the advantages of the proposed method. [ABSTRACT FROM AUTHOR]

Artificial neural networks (ANNs) have traditionally been seen as black-box models, because, although they are able to find “hidden” relations between inputs and outputs with a high approximation capacity, their structure seldom provides any insights on the structure of the functions being approximated. Several research papers have tried to debunk the black-box nature of ANNs, since it limits the potential use of ANNs in many research areas. This paper is framed in this context and proposes a methodology to determine the individual and collective effects of the input variables on the outputs for classification problems based on the ANOVA-functional decomposition. The method is applied after the training phase of the ANN and allows researchers to rank the input variables according to their importance in the variance of the ANN output. The computation of the sensitivity indices for product unit neural networks is straightforward as those indices can be calculated analytically by evaluating the integrals in the ANOVA decomposition. Unfortunately, the sensitivity indices associated with ANNs based on sigmoidal basis functions or radial basis functions cannot be calculated analytically. In this paper, the indices for those kinds of ANNs are proposed to be estimated by the (quasi-) Monte Carlo method. [ABSTRACT FROM PUBLISHER]

Recent reports show that projection neural networks with a low-dimensional state space can enhance computation speed obviously. This paper proposes two projection neural networks with reduced model dimension and complexity (RDPNNs) for solving nonlinear programming (NP) problems. Compared with existing projection neural networks for solving NP, the proposed two RDPNNs have a low-dimensional state space and low model complexity. Under the condition that the Hessian matrix of the associated Lagrangian function is positive semi-definite and positive definite at each Karush–Kuhn–Tucker point, the proposed two RDPNNs are proven to be globally stable in the sense of Lyapunov and converge globally to a point satisfying the reduced optimality condition of NP. Therefore, the proposed two RDPNNs are theoretically guaranteed to solve convex NP problems and a class of nonconvex NP problems. Computed results show that the proposed two RDPNNs have a faster computation speed than the existing projection neural networks for solving NP problems. [ABSTRACT FROM AUTHOR]

In this paper, we show that a one-layer feedforward neural network with exponential activation functions in the inner layer and logarithmic activation in the output neuron is a universal approximator of convex functions. Such a network represents a family of scaled log-sum exponential functions, here named log-sum-exp ($\mathrm {LSE}_{T}$). Under a suitable exponential transformation, the class of $\mathrm {LSE}_{T}$ functions maps to a family of generalized posynomials $\mathrm {GPOS}_{T}$ , which we similarly show to be universal approximators for log-log-convex functions. A key feature of an $\mathrm {LSE}_{T}$ network is that, once it is trained on data, the resulting model is convex in the variables, which makes it readily amenable to efficient design based on convex optimization. Similarly, once a $\mathrm {GPOS}_{T}$ model is trained on data, it yields a posynomial model that can be efficiently optimized with respect to its variables by using geometric programming (GP). The proposed methodology is illustrated by two numerical examples, in which, first, models are constructed from simulation data of the two physical processes (namely, the level of vibration in a vehicle suspension system, and the peak power generated by the combustion of propane), and then optimization-based design is performed on these models. [ABSTRACT FROM AUTHOR]

Smart grids, microgrids, and pure electric powertrains are the key technologies for achieving the expected goals concerning the restraint of CO2 emissions and global warming. In this context, an effective use of electrochemical energy storage systems (ESSs) is mandatory. In particular, accurate state of charge (SoC) estimations are helpful for improving the ESS performances. To this aim, developing accurate models of electrochemical cells is necessary for implementing effective SoC estimators. Therefore, a novel neural network modeling technique is proposed in this paper. The main contribution consists in the development of a white-box neural design that provides helpful insights into the cell physics, together with a powerful nonlinear approximation capability, and a flexible system identification procedure. In order to do that, the system equations of a white-box equivalent circuit model (ECM) have been combined with computational intelligence techniques by approximating each circuit element with a dedicated neural network. The model performances have been analyzed in terms of model accuracy, SoC estimation effectiveness, and computational cost over two realistic data sets. Moreover, the proposed model has been compared with a white-box ECM and a gray-box neural network model. The results prove that the proposed modeling technique is able to provide useful improvements in the SoC estimation task with a competing computational cost. [ABSTRACT FROM AUTHOR]

The volume, veracity, variability, and velocity of data produced from the ever increasing network of sensors connected to Internet pose challenges for power management, scalability, and sustainability of cloud computing infrastructure. Increasing the data processing capability of edge computing devices at lower power requirements can reduce several overheads for cloud computing solutions. This paper provides the review of neuromorphic CMOS-memristive architectures that can be integrated into edge computing devices. We discuss why the neuromorphic architectures are useful for edge devices and show the advantages, drawbacks, and open problems in the field of neuromemristive circuits for edge computing. [ABSTRACT FROM AUTHOR]

Due to object detection’s close relationship with video analysis and image understanding, it has attracted much research attention in recent years. Traditional object detection methods are built on handcrafted features and shallow trainable architectures. Their performance easily stagnates by constructing complex ensembles that combine multiple low-level image features with high-level context from object detectors and scene classifiers. With the rapid development in deep learning, more powerful tools, which are able to learn semantic, high-level, deeper features, are introduced to address the problems existing in traditional architectures. These models behave differently in network architecture, training strategy, and optimization function. In this paper, we provide a review of deep learning-based object detection frameworks. Our review begins with a brief introduction on the history of deep learning and its representative tool, namely, the convolutional neural network. Then, we focus on typical generic object detection architectures along with some modifications and useful tricks to improve detection performance further. As distinct specific detection tasks exhibit different characteristics, we also briefly survey several specific tasks, including salient object detection, face detection, and pedestrian detection. Experimental analyses are also provided to compare various methods and draw some meaningful conclusions. Finally, several promising directions and tasks are provided to serve as guidelines for future work in both object detection and relevant neural network-based learning systems. [ABSTRACT FROM AUTHOR]

This paper proposes a new and efficient technique to regularize the neural network in the context of deep learning using correlations among features. Previous studies have shown that oversized deep neural network models tend to produce a lot of redundant features that are either the shifted version of one another or are very similar and show little or no variations, thus resulting in redundant filtering. We propose a way to address this problem and show that such redundancy can be avoided using regularization and adaptive feature dropout mechanism. We show that regularizing both negative and positive correlated features according to their differentiation and based on their relative cosine distances yields network extracting dissimilar features with less overfitting and better generalization. This concept is illustrated with deep multilayer perceptron, convolutional neural network, sparse autoencoder, gated recurrent unit, and long short-term memory on MNIST digits recognition, CIFAR-10, ImageNet, and Stanford Natural Language Inference data sets. [ABSTRACT FROM AUTHOR]

This paper discusses the chaos prediction of nonlinear systems using various neural networks based on the modified substructure data-driven modeling architecture. In the modeling step, we construct two-coefficient loss functions according to the linear multi-step method to improve the prediction accuracy of neural networks. Then, the predicted response data of the system is given by the forward Euler method and neural networks. Under such architecture, chaos forecasting is carried out on a five-degree-of-freedom duffing oscillator system via the feedforward neural network (FNN), long short-term memory (LSTM) network and LSTM encoder-decoder (LSTM ED). The numerical simulation results show that the model can predict chaotic time series even if a small amount of information and samples are known, and the prediction window is twice that of the observation window. Among these models, LSTM ED exhibits the highest accuracy in both short-term and long-term chaos prediction. Furthermore, the prediction results mainly involve three evaluation indicators: absolute error, mean absolute error, normalized root mean square error. Through error analysis and noise processing, LSTM ED shows superior stability, robustness and extrapolation ability. Its prediction error is about half of FNN and the maximum increase in accuracy is 71.3 %. • Two-coefficient loss functions are constructed in the modeling step. • The forward Euler method is combined with neural networks in the prediction step. • Chaotic time series prediction based on various neural networks is investigated. • The accuracy, robustness and extrapolation ability of models are compared. [ABSTRACT FROM AUTHOR]

In this paper, we propose a general steganographic framework for neural networks to achieve covert communication. Firstly, we design a baseline steganographic method to embed secret data into convolutional layers of a given neural network (cover network) during the process of network training. With a network containing secret data (stego network), matrix multiplication is used to encode the parameters of convolutional layers for data extraction. Based on the baseline steganographic method, we respectively extended multi-source and multi-channel steganographic schemes for different scenarios, in which multiple agents are allowed to embed or extract secret data at the sender or receiver side. The multiple senders or receivers at the same side are ignorant of each other's existence, ensuring the independence and security of communication between each pair of sender and receiver. Experimental results verified the universality of the baseline method and the effectiveness of the extended schemes, including capacity, security, and robustness. [ABSTRACT FROM AUTHOR]

The softmax function is widely used in deep neural networks (DNNs), its hardware performance plays an important role in the training and inference of DNN accelerators. However, due to the complexity of the traditional softmax, the existing hardware architectures are resource-consuming or have low precision. In order to address the challenges, we study a base-2 softmax function in terms of its suitability for neural network training and efficient hardware implementation. Compared to the classical base- $e$ softmax function, the base-2 softmax function is a new softmax function that uses 2 as the exponential base instead of $e$. From the aspects of mathematical derivation and software simulation, we first demonstrate the feasibility and good accuracy of the base-2 softmax function in the application of neural network training. Then, we use the symmetric-mapping lookup table (SM-LUT) method to design a low-complexity architecture but with high precision to implement it. Under TSMC 28nm CMOS technology, an example design of our architecture has the area of $5676 ~\mu m^{2}$ and the power consumption of 13.12 mW for circuit synthesis at the frequency of 3 GHz. Compared with the latest works, our architecture achieves the best performance and efficiency. [ABSTRACT FROM AUTHOR]

Multi-bit-width convolutional neural network (CNN) maintains the balance between network accuracy and hardware efficiency, thus enlightening a promising method for accurate yet energy-efficient edge computing. In this work, we develop state-of-the-art multi-bit-width accelerator for NAS Optimized deep learning neural networks. To efficiently process the multi-bit-width network inferencing, multi-level optimizations have been proposed. Firstly, differential Neural Architecture Search (NAS) method is adopted for the high accuracy multi-bit-width network generation. Secondly, hybrid Booth based multi-bit-width multiply-add-accumulation (MAC) unit is developed for data processing. Thirdly, vector systolic array is proposed for effectively accelerating the matrix multiplications. With vector-style systolic dataflow, both the processing time and logic resources consumption can be reduced when compared with the classical systolic array. Finally, The proposed multi-bit-width CNN acceleration scheme has been practically deployed on FPGA platform of Xilinx ZCU102. Average performance on accelerating the full NAS optimized VGG16 network is 784.2 GOPS, and peek performance of the convolutional layer can reach as high as 871.26 GOPS for INT8, 1676.96 GOPS for INT4, and 2863.29 GOPS for INT2 respectively, which is among the best results in previous CNN accelerator benchmarks. [ABSTRACT FROM AUTHOR]

This paper presents a new study on the adaptive neural network-based control of a class of noncanonical nonlinear systems with large parametric uncertainties. Unlike commonly studied canonical form nonlinear systems whose neural network approximation system models have explicit relative degree structures, which can directly be used to derive parameterized controllers for adaptation, noncanonical form nonlinear systems usually do not have explicit relative degrees, and thus their approximation system models are also in noncanonical forms. It is well-known that the adaptive control of noncanonical form nonlinear systems involves the parameterization of system dynamics. As demonstrated in this paper, it is also the case for noncanonical neural network approximation system models. Effective control of such systems is an open research problem, especially in the presence of uncertain parameters. This paper shows that it is necessary to reparameterize such neural network system models for adaptive control design, and that such reparameterization can be realized using a relative degree formulation, a concept yet to be studied for general neural network system models. This paper then derives the parameterized controllers that guarantee closed-loop stability and asymptotic output tracking for noncanonical form neural network system models. An illustrative example is presented with the simulation results to demonstrate the control design procedure, and to verify the effectiveness of such a new design method. [ABSTRACT FROM AUTHOR]

The paper addresses convergence of solutions for a class of differential inclusions termed differential variational inequalities (DVIs). Each DVI describes the dynamics of a neural network (NN) evolving in a closed hypercube of \BBR ^n and defined by a continuously differentiable, cooperative and (possibly) nonreciprocal vector field f. The main result in the paper is that under a new condition on f, which is called strong Kamke-Müller (SKM) condition, the solution semiflow generated by the DVI is strongly order preserving (SOP) and hence it satisfies a Limit Set Dichotomy and enjoys generic convergence properties. A characterization of the SKM condition is given in terms of the interconnection properties of the Jacobian matrix of f. In the case where f is an affine, or a linear, vector field the considered DVIs include two relevant classes of NNs, namely, the linear systems operating on a closed hypercube, also known as linear systems in saturated mode (LSSMs), and the full-range (FR) model of cellular neural networks (CNNs). By applying the results to LSSMs it is obtained that any cooperative LSSM with a (possibly) nonsymmetric and fully interconnected matrix is generically convergent. Analogous results hold for FRCNNs. All the obtained convergence results hold in the general case where the DVIs, and the LSSMs and FRCNNs, possess multiple equilibrium points. [ABSTRACT FROM PUBLISHER]

This paper investigates drive-response synchronization of a class of neural networks with time-varying delays and discontinuous activations. Discontinuous state feedback controller and adaptive controller are designed such that the considered model can realize exponential complete synchronization. Moreover, the convergence rate is explicitly estimated when state feedback control is utilized. The obtained results are also applicable to neural networks with continuous activations since they are a special case of neural networks with discontinuous activations. Results of this paper improve corresponding ones which only quasi-synchronization can be achieved for neural networks with discontinuous activations. Finally, numerical simulations are given to verify the effectiveness of the theoretical results. [ABSTRACT FROM PUBLISHER]

This paper is concerned with the multistability of switched neural networks with piecewise linear activation functions under state-dependent switching. Under some reasonable assumptions on the switching threshold and activation functions, by using the state-space decomposition method, contraction mapping theorem, and strictly diagonally dominant matrix theory, we can characterize the number of equilibria as well as analyze the stability/instability of the equilibria. More interesting, we can find that the switching threshold plays an important role for stable equilibria in the unsaturation regions of activation functions, and the number of stable equilibria of an $n$ -neuron switched neural network with state-dependent parameters increases to $3^{n}$ from $2^{n}$ in the conventional one. Furthermore, for two-neuron switched neural networks, the precise attraction basin of each stable equilibrium point can be figured out, and its boundary is composed of the stable manifolds of unstable equilibrium points and the switching lines. Two simulation examples are discussed in detail to substantiate the effectiveness of the theoretical analysis. [ABSTRACT FROM AUTHOR]

Batch normalization (BN) has recently become a standard component for accelerating and improving the training of deep neural networks (DNNs). However, BN brings in additional calculations, consumes more memory, and significantly slows down the training iteration. Furthermore, the nonlinear square and sqrt operations in the normalization process impede low bit-width quantization techniques, which draw much attention to the deep learning hardware community. In this paper, we propose an $L1$ -norm BN (L1BN) with only linear operations in both forward and backward propagations during training. L1BN is approximately equivalent to the conventional $L2$ -norm BN (L2BN) by multiplying a scaling factor that equals $({\pi }/{2})^{1/2}$. Experiments on various convolutional neural networks and generative adversarial networks reveal that L1BN can maintain the same performance and convergence rate as L2BN but with higher computational efficiency. In real application-specified integrated circuit synthesis with reduced resources, L1BN achieves 25% speedup and 37% energy saving compared to the original L2BN. Our hardware-friendly normalization method not only surpasses L2BN in speed but also simplifies the design of deep learning accelerators. Last but not least, L1BN promises a fully quantized training of DNNs, which empowers future artificial intelligence applications on mobile devices with transfer and continual learning capability. [ABSTRACT FROM AUTHOR]

In this paper, the dissipative synchronization control problem for Markovian jump memristive neural networks (MNNs) is addressed with fully considering the time-varying delays and the fragility problem in the process of implementing the gain-scheduled controller. A Markov jump model is introduced to describe the stochastic changing among the connection of MNNs and it makes the networks under consideration suitable for some actual circumstances. By utilizing some improved integral inequalities and constructing a proper Lyapunov–Krasovskii functional, several delay-dependent synchronization criteria with less conservatism are established to ensure the dynamic error system is strictly stochastically dissipative. Based on these criteria, the procedure of designing the desired nonfragile gain-scheduled controller is established, which can well handle the fragility problem in the process of implementing the controller. Finally, an illustrated example is employed to explain that the developed method is efficient and available. [ABSTRACT FROM AUTHOR]

Online peer-to-peer (P2P) lending is expected to benefit both investors and borrowers due to their low transaction cost and the elimination of expensive intermediaries. From the lenders’ perspective, maximizing their return on investment is an ultimate goal during their decision-making procedure. In this paper, we explore and address a fundamental problem underlying such a goal: how to represent the two competing risks, charge-off and prepayment, in funded loans. We propose to model both potential risks simultaneously, which remains largely unexplored until now. We first develop a hierarchical grading framework to integrate two risks of loans both qualitatively and quantitatively. Afterward, we introduce an end-to-end deep learning approach to solve this problem by breaking it down into multiple binary classification subproblems that are amenable to both feature representation and risks learning. Particularly, we leverage deep neural networks to jointly solve these subtasks, which leads to the in-depth exploration of the interaction involved in these tasks. To the best of our knowledge, this is the first attempt to characterize competing risks for loans in P2P lending via deep neural networks. The comprehensive experiments on real-world loan data show that our methodology is able to achieve an appealing investment performance by modeling the competition within and between risks explicitly and properly. The feature analysis based on saliency maps provides useful insights into payment dynamics of loans for potential investors intuitively. [ABSTRACT FROM AUTHOR]

An efficient deep learning requires a memory-efficient construction of a neural network. This paper introduces a layerwise tensorized formulation of a multilayer neural network, called LTNN, such that the weight matrix can be significantly compressed during training. By reshaping the multilayer neural network weight matrix into a high-dimensional tensor with a low-rank approximation, significant network compression can be achieved with maintained accuracy. An according layerwise training is developed by a modified alternating least-squares method with backward propagation for fine-tuning only. LTNN can provide the state-of-the-art results on various benchmarks with significant compression. For MNIST benchmark, LTNN shows $64 \times $ compression rate without accuracy drop. For Imagenet12 benchmark, our proposed LTNN achieves $35.84 \times $ compression of the neural network with around 2% accuracy drop. We have also shown $1.615 \times $ faster on inference speed than the existing works due to the smaller tensor core ranks. [ABSTRACT FROM AUTHOR]

In this paper, a novel, automated process for constructing and initializing deep feedforward neural networks based on decision trees is presented. The proposed algorithm maps a collection of decision trees trained on the data into a collection of initialized neural networks with the structures of the networks determined by the structures of the trees. The tree-informed initialization acts as a warm-start to the neural network training process, resulting in efficiently trained, accurate networks. These models, referred to as “deep jointly informed neural networks” (DJINN), demonstrate high predictive performance for a variety of regression and classification data sets and display comparable performance to Bayesian hyperparameter optimization at a lower computational cost. By combining the user-friendly features of decision tree models with the flexibility and scalability of deep neural networks, DJINN is an attractive algorithm for training predictive models on a wide range of complex data sets. [ABSTRACT FROM AUTHOR]

Cross-modal hashing has attracted increasing research attention due to its efficiency for large-scale multimedia retrieval. With simultaneous feature representation and hash function learning, deep cross-modal hashing (DCMH) methods have shown superior performance. However, most existing methods on DCMH adopt binary quantization functions (e.g., $\text {sign}(\cdot)$) to generate hash codes, which limit the retrieval performance since binary quantization functions are sensitive to the variations of numeric values. Toward this end, we propose a novel end-to-end ranking-based hashing framework, in this paper, termed as deep semantic-preserving ordinal hashing (DSPOH), to learn hash functions with deep neural networks by exploring the ranking structure of feature dimensions. In DSPOH, the ordinal representation, which encodes the relative rank ordering of feature dimensions, is explored to generate hash codes. Such ordinal embedding benefits from the numeric stability of rank correlation measures. To make the hash codes discriminative, the ordinal representation is expected to well predict the class labels so that the ranking-based hash function learning is optimally compatible with the label predicting. Meanwhile, the intermodality similarity is preserved to guarantee that the hash codes of different modalities are consistent. Importantly, DSPOH can be effectively integrated with different types of network architectures, which demonstrates the flexibility and scalability of our proposed hashing framework. Extensive experiments on three widely used multimodal data sets show that DSPOH outperforms state of the art for cross-modal retrieval tasks. [ABSTRACT FROM AUTHOR]