Separate Deep Neural Networks Research Articles

Deep neural networks for explainable feature extraction in orchid identification

AbstractAutomated image-based plant identification systems are black-boxes, failing to provide an explanation of a classification. Such explanations are seen as being essential by taxonomists and are part of the traditional procedure of plant identification. In this paper, we propose a different method by extracting explicit features from flower images that can be employed to generate explanations. We take the benefit of feature extraction derived from the taxonomic characteristics of plants, with the orchids as an example domain. Feature classifiers were developed using deep neural networks. Two different methods were studied: (1) a separate deep neural network was trained for every individual feature, and (2) a single, multi-label, deep neural network was trained, combining all features. The feature classifiers were tested in predicting 63 orchid species using naive Bayes (NB) and tree-augmented Bayesian networks (TAN). The results show that the accuracy of the feature classifiers is in the range 83-93%. By combining these features using NB and TAN the species can be predicted with an accuracy of 88.9%, which is better than a standard pre-trained deep neural-network architecture, but inferior to a deep learning architecture after fine-tuning of multiple layers. The proposed novel feature extraction method still performs well for identification and is explainable, as opposed to black-box solutions that only aim for the best performance. Graphical abstract

Deep Learning-Based Network-Wide Energy Efficiency Optimization in Ultra-Dense Small Cell Networks

In ultra-dense small cell networks (UDSCNs), where a significant number of small cell base stations (SBSs) coexist, the amount of power consumed at the SBSs can be extremely high, rendering the efficient management of power consumption for the SBSs particularly important. Herein, we propose a deep-learning-based resource allocation strategy to maximize network-wide energy efficiency in the UDSCN by optimally controlling the transmit power and user association. In this regard, a novel deep neural network (DNN) structure comprising three separate DNN units, each of which determines the activation of the SBSs, user association, and transmit power, as well as an unsupervised-learning-based training methodology are designed. Simulation results verify that the proposed scheme achieves a near-optimal performance while requiring a short computation time.

RAID: Robust and Interpretable Daily Peak Load Forecasting via Multiple Deep Neural Networks and Shapley Values

Accurate daily peak load forecasting (DPLF) is crucial for informed decision-making in energy management. Deep neural networks (DNNs) are particularly apt for DPLF because they can analyze multiple factors, such as timestamps, weather conditions, and historical electric loads. Interpretability of machine learning models is essential for ensuring stakeholders understand and trust the decision-making process. We proposed the RAID (robust and interpretable DPLF) model, which enhances DPLF accuracy by recognizing daily peak load patterns and building separate DNN models for each day of the week. This approach was accessible for energy providers with limited computational resources, as the DNN models could be configured without a graphics processing unit (GPU). We utilized scikit-learn’s MLPRegressor for streamlined implementation, Optuna for hyperparameter optimization, and the Shapley additive explanations (SHAP) method to ensure interpretability. Applied to a dataset from two commercial office buildings in Richland, Washington, RAID outperformed existing methods like recurrent neural networks, Cubist, and HYTREM, achieving the lowest mean absolute percentage error values: 14.67% for Building 1 and 12.74% for Building 2. The kernel SHAP method revealed the influence of the previous day’s peak load and temperature-related variables on the prediction. The RAID model substantially improved energy management through enhanced DPLF accuracy, outperforming competing methods, providing a GPU-free configuration, and ensuring interpretable decision-making, with the potential to influence energy providers’ choices and promote overall energy system sustainability.

Computational microscopy for fast widefield deep-tissue fluorescence imaging using a commercial dual-cannula probe.

A solid-glass cannula serves as a micro-endoscope that can deliver excitation light deep inside tissue while also collecting emitted fluorescence. Then, we utilize deep neural networks to reconstruct images from the collected intensity distributions. By using a commercially available dual-cannula probe, and training a separate deep neural network for each cannula, we effectively double the field of view compared to prior work. We demonstrated ex vivo imaging of fluorescent beads and brain slices and in vivo imaging from whole brains. We clearly resolved 4 μm beads, with FOV from each cannula of 0.2 mm (diameter), and produced images from a depth of ~1.2 mm in the whole brain, currently limited primarily by the labeling. Since no scanning is required, fast widefield fluorescence imaging limited primarily by the brightness of the fluorophores, collection efficiency of our system, and the frame rate of the camera becomes possible.

Application of a Neural Network to Store and Compute the Optical Properties of Non-Spherical Particles

Radiative transfer simulations and remote sensing studies fundamentally require accurate and efficient computation of the optical properties of non-spherical particles. This paper proposes a deep learning (DL) scheme in conjunction with an optical property database to achieve this goal. Deep neural network (DNN) architectures were obtained from a dataset of the optical properties of super-spheroids with extensive shape parameters, size parameters, and refractive indices. The dataset was computed through the invariant imbedding T-matrix method. Four separate DNN architectures were created to compute the extinction efficiency factor, single-scattering albedo, asymmetry factor, and phase matrix. The criterion for designing these neural networks was the achievement of the highest prediction accuracy with minimal DNN parameters. The numerical results demonstrate that the determination coefficients are greater than 0.999 between the prediction values from the neural networks and the truth values from the database, which indicates that the DNN can reproduce the optical properties in the dataset with high accuracy. In addition, the DNN model can robustly predict the optical properties of particles with high accuracy for shape parameters or refractive indices that are unavailable in the database. Importantly, the ratio of the database size (∼127 GB) to that of the DNN parameters (∼20 MB) is approximately 6810, implying that the DNN model can be treated as a highly compressed database that can be used as an alternative to the original database for real-time computing of the optical properties of non-spherical particles in radiative transfer and atmospheric models.

Deep Learning-Assisted Real-Time Forward Modeling of Electromagnetic Logging in Complex Formations

Higher dimensional (i.e., 2-D and 3-D) modeling is indispensable to correctly evaluate the responses of electromagnetic (EM) logging tools in complex formation environments. However, limited by the high computational cost of rigorous modeling, such as the finite-difference method and the finite-element method, the real-time applications in the well logging industry primarily rely on the 1-D forward solver, which would result in erroneous formation evaluation for complex scenarios. As a result, aiming at realizing fast modeling for EM logging tools in complex formations, this letter proposes a general framework assisted by deep neural networks (DNNs). The framework consists of three modules: earth model classification, parameter extraction, and surrogate construction. Separate DNNs are trained and tested for different modules. The accuracy and efficiency of the DNN-assisted fast modeling are validated by several experiments. This study finds that the fast modeling assisted by DNNs is able to calculate the tool responses and reconstruct the subsurface formations in real time.

Modeling Liquid Water by Climbing up Jacob's Ladder in Density Functional Theory Facilitated by Using Deep Neural Network Potentials.

Within the framework of Kohn-Sham density functional theory (DFT), the ability to provide good predictions of water properties by employing a strongly constrained and appropriately normed (SCAN) functional has been extensively demonstrated in recent years. Here, we further advance the modeling of water by building a more accurate model on the fourth rung of Jacob's ladder with the hybrid functional, SCAN0. In particular, we carry out both classical and Feynman path-integral molecular dynamics calculations of water with the SCAN0 functional and the isobaric-isothermal ensemble. To generate the equilibrated structure of water, a deep neural network potential is trained from the atomic potential energy surface based on ab initio data obtained from SCAN0 DFT calculations. For the electronic properties of water, a separate deep neural network potential is trained by using the Deep Wannier method based on the maximally localized Wannier functions of the equilibrated trajectory at the SCAN0 level. The structural, dynamic, and electric properties of water were analyzed. The hydrogen-bond structures, density, infrared spectra, diffusion coefficients, and dielectric constants of water, in the electronic ground state, are computed by using a large simulation box and long simulation time. For the properties involving electronic excitations, we apply the GW approximation within many-body perturbation theory to calculate the quasiparticle density of states and bandgap of water. Compared to the SCAN functional, mixing exact exchange mitigates the self-interaction error in the meta-generalized-gradient approximation and further softens liquid water toward the experimental direction. For most of the water properties, the SCAN0 functional shows a systematic improvement over the SCAN functional. However, some important discrepancies remain. The H-bond network predicted by the SCAN0 functional is still slightly overstructured compared to the experimental results.

Ensemble diversified learning for image classification with noisy labels

In this work, we develop a new approach for learning a deep neural network for image classification with noisy labels using ensemble diversified learning. We first partition the training set into multiple subsets with diversified image characteristics. For each subset, we train a separate deep neural network image classifier. These networks are then used to encode the input image into different feature vectors, providing diversified observations of the input image. The encoded features are then fused together and further analyzed by a decision network to produce the final classification output. We study image classification on noisy labels with and without the access to clean samples. Our extensive experimental results on the CIFAR-10 and MNIST datasets demonstrate that our proposed method outperforms existing methods by a large margin.

Target-Dependent Scalable Image Compression Using a Reconfigurable Recurrent Neural Network

Conventional human-centric image compression techniques are optimized for human visual perception, and are generally evaluated by metrics such as MSSSIM and PSNR. On the other hand, task-centric image compression techniques that target deep neural networks (DNN) inference focus on understanding images, and are measured by inference accuracy. As images should be encoded and decoded differently depending on the target metric, existing deep learning based compression techniques have focused on designing an independent neural network that is optimized for either one of the two targets. However, as these techniques require the target metric to be determined during a training phase, separate DNN models should be trained for different target applications, allowing no scalability between human- and task-centric compression. In this paper, we propose a target-dependent scalable image compression pipeline, where compressed images can be differently decoded in real time depending on the interpolated target between perceptual quality and DNN inference accuracy. This dynamic scalability is supported by a reconfigurable recurrent neural network (RNN) that can dynamically change its flow according to the interpolation parameter. The experimental results show that the proposed method achieves comparable MSSSIM and classification accuracy to the compression networks individually trained for human-centric and task-centric targets, respectively, while providing scalability between the two targets depending on the interpolation parameter.

Integrated Replay Spoofing-Aware Text-Independent Speaker Verification

A number of studies have successfully developed speaker verification or presentation attack detection systems. However, studies integrating the two tasks remain in the preliminary stages. In this paper, we propose two approaches for building an integrated system of speaker verification and presentation attack detection: an end-to-end monolithic approach and a back-end modular approach. The first approach simultaneously trains speaker identification, presentation attack detection, and the integrated system using multi-task learning using a common feature. However, through experiments, we hypothesize that the information required for performing speaker verification and presentation attack detection might differ because speaker verification systems try to remove device-specific information from speaker embeddings, while presentation attack detection systems exploit such information. Therefore, we propose a back-end modular approach using a separate deep neural network (DNN) for speaker verification and presentation attack detection. This approach has thee input components: two speaker embeddings (for enrollment and test each) and prediction of presentation attacks. Experiments are conducted using the ASVspoof 2017-v2 dataset, which includes official trials on the integration of speaker verification and presentation attack detection. The proposed back-end approach demonstrates a relative improvement of 21.77% in terms of the equal error rate for integrated trials compared to a conventional speaker verification system.

A hybrid speech enhancement system with DNN based speech reconstruction and Kalman filtering

In this paper, we propose a hybrid speech enhancement system that exploits deep neural network (DNN) for speech reconstruction and Kalman filtering for further denoising, with the aim to improve performance under unseen noise conditions. Firstly, two separate DNNs are trained to learn the mapping from noisy acoustic features to the clean speech magnitudes and line spectrum frequencies (LSFs), respectively. Then the estimated clean magnitudes are combined with the phase of the noisy speech to reconstruct the estimated clean speech, while the LSFs are converted to linear prediction coefficients (LPCs) to implement Kalman filtering. Finally, the reconstructed speech is Kalman-filtered for further removing the residual noises. The proposed hybrid system takes advantage of both the DNN based reconstruction and traditional Kalman filtering, and can work reliably in either matched or unmatched acoustic environments. Computer based experiments are conducted to evaluate the proposed hybrid system with comparison to traditional iterative Kalman filtering and several state-of-the-art DNN based methods under both seen and unseen noises. It is shown that compared to the DNN based methods, the hybrid system achieves similar performance under seen noise, but notably better performance under unseen noise, in terms of both speech quality and intelligibility.

A Multimodal Feature Fusion-Based Deep Learning Method for Online Fault Diagnosis of Rotating Machinery.

Rotating machinery usually suffers from a type of fault, where the fault feature extracted in the frequency domain is significant, while the fault feature extracted in the time domain is insignificant. For this type of fault, a deep learning-based fault diagnosis method developed in the frequency domain can reach high accuracy performance without real-time performance, whereas a deep learning-based fault diagnosis method developed in the time domain obtains real-time diagnosis with lower diagnosis accuracy. In this paper, a multimodal feature fusion-based deep learning method for accurate and real-time online diagnosis of rotating machinery is proposed. The proposed method can directly extract the potential frequency of abnormal features involved in the time domain data. Firstly, multimodal features corresponding to the original data, the slope data, and the curvature data are firstly extracted by three separate deep neural networks. Then, a multimodal feature fusion is developed to obtain a new fused feature that can characterize the potential frequency feature involved in the time domain data. Lastly, the fused new feature is used as the input of the Softmax classifier to achieve a real-time online diagnosis result from the frequency-type fault data. A simulation experiment and a case study of the bearing fault diagnosis confirm the high efficiency of the method proposed in this paper.

DNN-based speech enhancement using soft audible noise masking for wind noise reduction

This paper presents a deep neural network (DNN)-based speech enhancement algorithm based on the soft audible noise masking for the single-channel wind noise reduction. To reduce the low-frequency residual noise, the psychoacoustic model is adopted to calculate the masking threshold from the estimated clean speech spectrum. The gain for noise suppression is obtained based on soft audible noise masking by comparing the estimated wind noise spectrum with the masking threshold. To deal with the abruptly time-varying noisy signals, two separate DNN models are utilized to estimate the spectra of clean speech and wind noise components. Experimental results on the subjective and objective quality tests show that the proposed algorithm achieves the better performance compared with the conventional DNN-based wind noise reduction method.

Detecting 3D Points of Interest Using Multiple Features and Stacked Auto-encoder.

Considering the fact that points of interest on 3D shapes can be discriminated from a geometric perspective, it is reasonable to map the geometric signature of a point $p$p to a probability value encoding to what degree $p$p is a point of interest, especially for a specific class of 3D shapes. Based on the observation, we propose a three-phase algorithm for learning and predicting points of interest on 3D shapes by using multiple feature descriptors. Our algorithm requires two separate deep neural networks (stacked auto-encoders) to accomplish the task. During the first phase, we predict the membership of the given 3D shape according to a set of geometric descriptors using a deep neural network. After that, we train the other deep neural network to predict a probability distribution defined on the surface representing the possibility of a point being a point of interest. Finally, we use a manifold clustering technique to extract a set of points of interest as the output. Experimental results show superior detection performance of the proposed method over the previous state-of-the-art approaches.

Arnold: An Autonomous Agent to Play FPS Games

Advances in deep reinforcement learning have allowed autonomous agents to perform well on Atari games, often outperforming humans, using only raw pixels to make their decisions. However, most of these games take place in 2D environments that are fully observable to the agent. In this paper, we present Arnold, a completely autonomous agent to play First-Person Shooter Games using only screen pixel data and demonstrate its effectiveness on Doom, a classical first-person shooter game. Arnold is trained with deep reinforcement learning using a recent Action-Navigation architecture, which uses separate deep neural networks for exploring the map and fighting enemies. Furthermore, it utilizes a lot of techniques such as augmenting high-level game features, reward shaping and sequential updates for efficient training and effective performance. Arnold outperforms average humans as well as in-built game bots on different variations of the deathmatch. It also obtained the highest kill-to-death ratio in both the tracks of the Visual Doom AI Competition and placed second in terms of the number of frags.

Deep-neural network approaches for speech recognition with heterogeneous groups of speakers including children

AbstractThis paper introduces deep neural network (DNN)–hidden Markov model (HMM)-based methods to tackle speech recognition in heterogeneous groups of speakers including children. We target three speaker groups consisting of children, adult males and adult females. Two different kind of approaches are introduced here: approaches based on DNN adaptation and approaches relying on vocal-tract length normalisation (VTLN). First, the recent approach that consists in adapting a general DNN to domain/language specific data is extended to target age/gender groups in the context of DNN–HMM. Then, VTLN is investigated by training a DNN–HMM system by using either mel frequency cepstral coefficients normalised with standard VTLN or mel frequency cepstral coefficients derived acoustic features combined with the posterior probabilities of the VTLN warping factors. In this later, novel, approach the posterior probabilities of the warping factors are obtained with a separate DNN and the decoding can be operated in a single pass when the VTLN approach requires two decoding passes. Finally, the different approaches presented here are combined to take advantage of their complementarity. The combination of several approaches is shown to improve the baseline phone error rate performance by thirty per cent to thirty-five per cent relative and the baseline word error rate performance by about ten per cent relative.

Ensemble of deep neural networks using acoustic environment classification for statistical model-based voice activity detection

In this paper, we investigate the ensemble of deep neural networks (DNNs) by using an acoustic environment classification (AEC) technique for the statistical model-based voice activity detection (VAD). From an investigation of the statistical model-based VAD, it is known that the traditional decision rule is based on the geometric mean of the likelihood ratio or the support vector machine (SVM), which is a shallow model with zero or one hidden layer. Since the shallow models cannot take an advantage of the diversity of the space distribution of features, in the training step, we basically build the multiple DNNs according the different noise types by employing the parameters of the statistical model-based VAD algorithm. In addition, the separate DNN is designed for the AEC algorithm in order to choose the best DNN for each noise. In the on-line noise-aware VAD step, the AEC is first performed on a frame-by-frame basis using the separate DNN so the a posteriori probabilities to identify noise are obtained. Once the probabilities are achieved for each noise, the environmental knowledge is contributed to allow us to combine the speech presence probabilities which are derived from the ensemble of the DNNs trained for the individual noise. Our approach for VAD was evaluated in terms of objective measures and showed significant improvement compared to the conventional algorithm.

Deep Neural Networks for Single-Channel Multi-Talker Speech Recognition

We investigate techniques based on deep neural networks (DNNs) for attacking the single-channel multi-talker speech recognition problem. Our proposed approach contains five key ingredients: a multi-style training strategy on artificially mixed speech data, a separate DNN to estimate senone posterior probabilities of the louder and softer speakers at each frame, a weighted finite-state transducer (WFST)-based two-talker decoder to jointly estimate and correlate the speaker and speech, a speaker switching penalty estimated from the energy pattern change in the mixed-speech, and a confidence based system combination strategy. Experiments on the 2006 speech separation and recognition challenge task demonstrate that our proposed DNN-based system has remarkable noise robustness to the interference of a competing speaker. The best setup of our proposed systems achieves an average word error rate (WER) of 18.8% across different SNRs and outperforms the state-of-the-art IBM superhuman system by 2.8% absolute with fewer assumptions.