Channel Cross Research Articles

A flexible adhesive surface electrode array capable of cervical electroneurography during a sequential autonomic stress challenge

This study introduces a flexible, adhesive-integrated electrode array that was developed to enable non-invasive monitoring of cervical nerve activity. The device uses silver-silver chloride as the electrode material of choice and combines it with an electrode array consisting of a customized biopotential data acquisition unit and integrated graphical user interface (GUI) for visualization of real-time monitoring. Preliminary testing demonstrated this electrode design can achieve a high signal to noise ratio during cervical neural recordings. To demonstrate the capability of the surface electrodes to detect changes in cervical neuronal activity, the cold-pressor test (CPT) and a timed respiratory challenge were employed as stressors to the autonomic nervous system. This sensor system recording, a new technique, was termed Cervical Electroneurography (CEN). By applying a custom spike sorting algorithm to the electrode measurements, neural activity was classified in two ways: (1) pre-to-post CPT, and (2) during a timed respiratory challenge. Unique to this work: (1) rostral to caudal channel position-specific (cephalad to caudal) firing patterns and (2) cross challenge biotype-specific change in average CEN firing, were observed with both CPT and the timed respiratory challenge. Future work is planned to develop an ambulatory CEN recording device that could provide immediate notification of autonomic nervous system activity changes that might indicate autonomic dysregulation in healthy subjects and clinical disease states.

Shape-Enhanced Open-Channel Hydrodynamic Chromatography.

Hydrodynamic chromatography (HDC) is a well-established analytical separation method for the size separation of nano- and microparticles and large molecular weight solutes such as synthetic polymers and proteins. We report on a theoretical study showing that the separation resolution of open-tubular HDC can be significantly enhanced by changing the cross-sectional shape of the separation channel. By enforcing Brenner's macro-transport approach, we provide theoretical/numerical evidence showing how the shape of the cross section influences quantitatively both the selectivity and the axial dispersion of the suspended particles in HDC. The separation performance of square-, triangle-, and star-shaped channel cross sections is compared to that of a cylindrical capillary over three decades of the particle Péclet number in terms of the minimal separation length and time to obtain the unit resolution of a two-particle mixture. Enhancement factors up to 400% are found in the case of triangular shapes, with the best performing case being the 70.6° angle, which can be obtained by KOH etching of bulk silicon.

A new approach for accurate determination of particle sizes in microfluidic impedance cytometry

In microfluidic impedance cytometry, the change in impedance is recorded as an individual cell passes through a channel between electrodes deposited on its walls, and the particle size is inferred from the amplitude of the impedance signal using calibration. However, because the current density is nonuniform between electrodes of finite width, there could be an error in the particle size measurement because of uncertainty about the location of the particle in the channel cross section. Here, a correlation is developed relating the particle size to the signal amplitude and the velocity of the particle through the channel. The latter is inferred from the time interval between the two extrema in the impedance curve as the particle passes through a channel with cross-sectional dimensions of 50 μm (width) × 30 μm (height) with two pairs of parallel facing electrodes. The change in impedance is predicted using 3D COMSOL finite-element simulations, and a theoretical correlation that is independent of particle size is formulated to correct the particle diameter for variations in the cross-sectional location. With this correlation, the standard deviation in the experimental data is reduced by a factor of two to close to the standard deviation reported in the manufacturer specifications.

Effect of aspect ratio on entrance length in rectangular minichannels with plenum

In the case of a long, straight rectangular channel, hydrodynamic development of flows is influenced by the growth of the boundary layer along the walls of the channel. Though such a geometry is well-studied in the literature, in reality, the flow often happens in channels with plenums on each end and is not studied extensively. This work addresses this gap. There is a sudden contraction from the plenum to the channel which causes the flow to separate at the entrance of the channel. Hence, the flow development is influenced not only by the boundary layer growth but also by recirculation and the presence of a continuous wall along one direction in the case of planar geometries. This causes the centerline velocity in the entrance region to overshoot the value at the fully developed region, which makes the conventional usage of 99% of the fully developed value difficult. Hence, an alternate method of defining entrance length, based on the slowest development across the channel cross section, is proposed. Based on this approach, the entrance length value shows a non-monotonic variation with the aspect ratio (AR)—its value reduces between 0.6 and 1.66; beyond 1.66, it increases up to 20 before becoming flat. The entrance length also shows a weak dependence on the Reynolds number for AR between 2 and 20. A new set of correlations of entrance and recirculation lengths are proposed.

Thermal characteristics of refrigerant flow boiling in two mini-channel heat sinks of different aspect ratios for battery thermal management

The two-phase refrigerant cooling system has excellent thermal management performance, which can provide good temperature control of lithium-ion batteries. In this paper, the thermal performance of mini-channel heat sinks with different aspect ratios was investigated for a battery module with a small-scale vapor compression refrigeration system. Two mini-channel heat sinks (MiCHSs), one with a narrow channel cross section of 2 mm × 4.5 mm (MiCHS1) and the other one with square channel cross section of 3 mm × 3 mm (MiCHS2), were fabricated with a visualization window and connected in parallel in the system. The experiments were run at different openings of expansion valve (OEV) and heating power. With the increase of OEV, the refrigerant subjected to fixed heating power in the mini-channels experiences different flow patterns of annular flow, slug flow and small-bubble flow successively. The best cooling performance can be achieved in slug flow, corresponding to an optimal OEV. The optimal OEV value increased from 32 % to 55 % with the increase of heating power ranging from 100 W to 200 W. Under the same optimal OEV condition, the heat transfer coefficient and average temperature of MiCHSs increase with the increase of heating power for both MiCHSs. The channel aspect ratio has an effect on the heat transfer performance of MiCHS. Compared with the narrow channels of MiCHS 1 with a large aspect ratio, the flow boiling in square channel of MiCHS2 can achieve a better cooling performance due to the well-confined bubbly flow and thinner liquid film on the base of the mini-channels.

Classification from Sky: A Robust Remote Sensing Time Series Image Classification Using Spatial Encoder and Multi-Fast Channel Attention

The unparalleled availability of Satellite Image Time Series (SITS) for crop phenology classification unravels agricultural parcel observation and monitoring with applications of both economic and ecological importance. Moreover, the need for distinct classification of agricultural parcels into individual crop types falls on state-of-the-art deep learning models for this extrinsic task. However, most existing approaches implemented are complex and ineffective attention incorporated models, which in turn lack the resilience to recognize useful bands in achieving greater accuracy. We propose a Multi-Fast Channel Attention module for deep CNNs based on a Spatial Encoder (SE-MFCA) that requires a few parameters while enhancing the performance-complexity trade-off dilemma. Hence, we leverage on spatial encoder module to extract the images as disorderly sets of pixels to enhance the coarse spatial resolution features. We empirically show that appropriate parameter sharing in the cross channel interaction can preserve performance while significantly reducing model complexity. The proposed multi-channel attention module can efficiently be implemented via an encoder-decoder network to prevent the loss of detailed spatial information. Again, we parallelly distributed the input channel into multiple heads in our network to recover the specialized input features, which will concatenate with the residual to form a rich single feature representation. The extensive experiment has shown that our model SE-MFCA is efficient and effective compared with the previous state-of-the-art time series classification algorithm on a publicly available dataset of Sentinel-2 images for agricultural parcels. Performance-wise SE-MFCA achieves the highest overall accuracy of 94.50% and the highest mean intersection over union score of 51.92%, besides the least trainable params of 131 K and fewer floating point operations of 0.16 M.

Conventional and Cross-Channel MR Radiomic Features do Not Predict Histone H3 Status in DIPG: Genomic and Clinical Evaluation of a Multi-Institutional Cohort

<h3>Purpose/Objective(s)</h3> Genomic profiling of diffuse pontine glioma (DIPG) has suggested distinct and clinically relevant molecular subgroups based on the presence and type of histone H3K27M mutation. Recent studies have demonstrated improved safety for biopsy, but sampling carries risks. We evaluated the impact of radiomic features on the molecular classification of 154 histologically confirmed and centrally reviewed DIPG. <h3>Materials/Methods</h3> Automatically segmented and manually reviewed tumor volumes were used to compute single & cross channel radiomic features from T1, T2, FLAIR, and T1+gadolinium MR sequences obtained from histone H3 WT (N=30) and H3K27M-mutant (N=124) DIPG. Radiomic feature stability was assessed when multiple pretreatment MRs were available using the concordance correlation coefficient. Random Forest, Logistic Regression, XGBoost, and Deep Learning classifiers were used to build predictive models of H3K27M status in a training dataset, after exclusion of correlated features. Model accuracy was assessed in the remaining validation dataset. <h3>Results</h3> Two hundred and twelve of 456 imaging features (46.4%) demonstrated stability (CCC, p<0.05) between pretreatment scans. Accuracy varied by method, but none surpassed the No Information Rate (82%). Model accuracy was not improved by efforts to remedy class imbalance (resampling and class weight adjustment). Several cross channel radiomic features were consistently selected by the models, indicating some predictive utility, although cross institutional scanning techniques were confounding. <h3>Conclusion</h3> While early efforts with single institutional data were encouraging, differences in MR technique and class imbalance seemed to limit the identification of stable radiomic features predictive of H3 status in this multi-institutional cohort. Biopsy remains key for determining trial eligibility and directing precision medicine.

Suppression of bacterial rheotaxis in wavy channels

Controlling the swimming behavior of bacteria is crucial, for example, to prevent contamination of ducts and catheters. We show that bacteria modeled by deformable microswimmers can accumulate in flows through straight microchannels either in their center or on previously unknown attractors near the channel walls. In flows through wavy microchannels we predict a resonance effect for semiflexible microswimmers. As a result, microswimmers can be deflected in a controlled manner so that they swim in modulated channels distributed over the channel cross section rather than localized near the wall or the channel center. Thus, depending on the flow amplitude, both upstream orientation of swimmers and their accumulation at the boundaries, which can promote surface rheotaxis, are suppressed. Our results suggest strategies for controlling the behavior of live and synthetic swimmers in microchannels.

Machine-learning-aided design optimization of internal flow channel cross-sections

• Artificial neural network-based models are developed for prediction of nusselt number and friction factor. • The predictions apply for fully developed laminar flow in constant cross-section ducts of different cross section shapes. • The machine learning (ML) model predictions are validated and compared against numerical simulations. • Use of ML models in shape optimization of channel cross section for different objective functions is demonstrated. • Guidelines for usage of the developed ML models are discussed. The design optimization of various thermal management components such as cold plates, heat sinks, and heat exchangers relies on accurate prediction of flow heat transfer and pressure drop. During the iterative design process, the heat transfer and pressure drop is typically either computed numerically or obtained using geometry-specific correlations for Nusselt number and friction factor. Numerical approaches are accurate for evaluation of a single design but become computationally expensive if many design iterations are required (such as during formal optimization processes). Correlation-based approaches restrict the design space to a specific set of geometries for which correlations are available. Surrogate models for the Nusselt number and friction factor, which are more universally applicable to all geometries than traditional correlations, would enable flexible and computationally inexpensive design optimization. The current work develops machine-learning-based surrogate models for predicting the Nusselt number and friction factor under fully developed internal flow in channels of arbitrary cross section and demonstrates use of these models for optimization of the cross-sectional channel shape. The predictive performance and generality of the machine learning surrogate models is first verified on various shapes outside the training dataset, and then the models are used in the design optimization of flow cross sections based on performance metrics that weigh both heat transfer and pressure drop. The optimization process leads to novel shapes outside the training data, and so numerical simulations are carried out on these optimized shapes to compare with the surrogate model predictions and show their performance is at least as good as that of shapes with known correlations available. A three-lobed shape was found to reduce friction factor, whereas a pentagon with rounded corners and an ice cream cone-shaped duct, both found using different performance metrics. Although the ML model predictions lose accuracy outside the training set for these novel shapes, the predictions follow the correct trends with parametric variations of the shape and therefore successfully direct the search toward optimized shapes.

A traffic state recognition model based on feature map and deep learning

Real-time and accurate traffic state identification can provide reference for urban traffic control and guidance. Due to the randomness and complexity of traffic flow, it is difficult to identify the traffic state accurately. The existing researches mainly adopted traffic state feature vectors and machine learning to identify the traffic state. Few studies attempted to use the feature map and deep learning for traffic state recognition. Therefore, this paper proposes a traffic state recognition model based on the traffic state feature map and deep learning. The feature map is a chromatogram form of digital traffic state feature vector. And the deep learning has strong predictive performance in image identification. In the model, the road traffic state feature vector is extracted from the vehicle trajectory data, and the Gram Angle Field (GAF) is adopted to transform the feature vector into feature map. Then, the deep learning algorithm is utilized in traffic state offline classification and online recognition. In the offline training phase, the historical vehicle trajectory data and the DeepCluster algorithm are used to establish the mapping relationship between the feature maps and the traffic states. In the online identification stage, the real-time vehicle trajectory data and CoAtNet algorithm are utilized to identify the current traffic state. In experiments, the proposed model is compared with the Convolutional Neural Network (CNN), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) models using the open vehicle trajectory data of Shanghai North Cross Channel. The results showed that the proposed model achieved the best performance with the accuracy of 92.06% and could provide support for road traffic control and guidance.

Experimental studies of a novel one-dimensional plasma window

The plasma window is a windowless vacuum sealing device that utilizes a cascade arc discharge to isolate regions with different pressures. It was proposed as an alternative to the traditional windowless sealing technology of differential pumping, which greatly improves the sealing efficiency and has been successfully applied in electron beam welding. The discharge channel of the existing plasma window is a small circular hole, referred to as a zero-dimensional plasma window, which has limited applications to its dimensions. In this study, we suggested a one-dimensional (1D) plasma window with a slit discharge channel (cross section: 3 × 35 mm2, length: 60 mm). An arc discharge is realized with the support of an 80 A DC power supply, which can maintain a pressure drop of more than 10 times between the high-pressure and low-pressure ends of the plasma window. A COMSOL-based magneto-hydrodynamic model of the plasma window was established, and simulations were in good agreement with the experimental results. The pressure drop in the plasma window caused by argon plasma has also been theoretically analyzed and discussed. The feasibility of a 1D plasma window to achieve vacuum isolation was verified theoretically, numerically, and experimentally.

Impact of Process-Induced Inclined Sidewalls on Gate-Induced Drain Leakage (GIDL) Current of Nanowire GAA MOSFETs

The shape of the channel cross section in rectangular nanowire (NW) gate-all-around (GAA) MOSFETs turns trapezoidal due to process variations. In this article, the impact of process-induced inclination of sidewalls on gate-induced drain leakage (GIDL) current in the trapezoidal channel NW GAA MOSFETs has been systematically investigated using experimental and calibrated TCAD simulation results. The GIDL current has also been analyzed against the variation in other device parameters, such as channel length, height, and width. The lateral band-to-band tunneling (L-BTBT) mechanism at the channel/drain junction has been considered in simulations to obtain the GIDL current. The investigation reveals that the GIDL current increases up to two times if the process-induced sidewalls inclination angle increases from 0° to 20°.

PIV-based experiments on density-induced flow in an estuarine navigation channel constricted by a groin

Density-induced flow caused by horizontal salinity gradients is an important factor affecting sediment transport in estuarine navigation channels. Groins are common engineering structures that are often used to regulate and narrow these channels. In this study, PIV based laboratory experiments were conducted to investigate the effects of the groin on density-induced flow in an estuarine navigation channel. Flow fields and turbulence characteristics of the density-induced currents were investigated for various horizontal salinity gradients and different lengths of groins. PIV observations clearly show vertically stratified density currents caused by salinity gradient. Results show that groins modify the velocity field in the channel, and density currents spread horizontally after passing through channel cross sections constricted by groins. PIV experimental analyses show that, in the narrow channel cross section, the average horizontal velocity is almost unchanged under different lengths of groins if the horizontal salinity-gradient force remains the same. Results also show that the density-induced flowrate reduces after the channel is narrowed by the groin. This indicates that groins can be used to reduce the density-induced flowrate and hence reduce sediment transport along with flow in an estuarine navigation channel.

Silicon Dielectric Diplexer Module for 600-GHz-Band Frequency-Division Multiplexing Wireless Communication

We developed a silicon (Si) dielectric diplexer module for 600-GHz-band frequency-division multiplexing wireless communication. This diplexer was developed based on Si dielectric waveguide technology, which provides low-loss transmission characteristics compared with the conventional metallic terahertz-transmission platform, such as metallic hollow waveguides. It includes straight waveguides based on an effective medium composed of Si and air, and a coupler comprising of two adjacent unclad Si waveguides. Furthermore, for the communication system integration, we developed a practical and low-insertion loss module with WR1.5 metallic hollow waveguide input–output interfaces and modularized the diplexer. We performed transmission measurement experiments with the fabricated diplexer module and achieved 3-dB bandwidths of 101 and 37 GHz, and crosstalk of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$-$</tex-math></inline-formula> 34 dB at 680 GHz and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$-$</tex-math></inline-formula> 41 dB at 618 GHz in the cross and bar directions, respectively. Finally, we performed a two-channel wireless communication experiment using the developed diplexer module in the 600-GHz band. We achieved error-free communication at 10 and 6 Gbit/s on the cross and bar channels, respectively. In addition, we achieved a data rate of up to 12.5 Gbit/s at the forward error correction limits for both channels.

Numerical simulations of cell sorting through inertial microfluidics

The dynamics of a cell suspended in a Newtonian liquid subjected to a pressure-driven flow at non-negligible inertia in cylindrical and square cross section microfluidic channels is studied through three-dimensional arbitrary Lagrangian–Eulerian finite-element numerical simulations. The cell is modeled through the neo-Hookean hyper-elastic constitutive equation, which can describe biological particles undergoing moderate deformations. The cell-to-channel relative dimension is fixed to 0.2, whereas the Reynolds number Re, measuring the relative importance of liquid inertial and viscous forces, and the elastic capillary number Cae, measuring the relative importance of liquid viscous stress and solid elastic stress, are varied by several orders of magnitude. In a cylindrical tube, the cell migrates transversally to the flow direction until reaching a radial equilibrium position depending on Re and Cae. Given Re, the softer the cell (i.e., the larger Cae) the closer its equilibrium position to the tube axis, thus allowing for the separation of healthy and diseased cells which have similar dimensions but different mechanical properties. In a channel with a square cross section, a much more complex dynamics is found. Depending on Re and Cae, the cell can either migrate to the channel centerline, to the closest median of the channel cross section (thus, four equilibrium positions can be identified due to symmetry), to the closest diagonal (again, four equilibrium positions), or to an intermediate position in between the median and the diagonal (eight equilibrium positions).

50-Fold Reduction of Separation Time in Open-Channel Hydrodynamic Chromatography via Lateral Vortices.

Despite its relatively long history, open-channel hydrodynamic chromatography (OC-HDC) still represents a niche technique for determining the size distribution of particle suspensions. Practical limitations of this separation method ultimately arise from the low eluent velocity that is necessary to contain the adverse increase of analyte bandwidth caused by Taylor-Aris dispersion. Because of the micrometric size of the channel cross section, the low eluent velocity translates into order of pL-per-minute flow rates, which introduce a challenge for both the injection and the detection systems. In this article, we provide theoretical/numerical evidence illustrating how a sizable reduction of the Taylor-Aris effect can be obtained by triggering cross-sectional vortices alongside the main pressure-driven axial flow. As a case study, we consider a square channel geometry where the lateral vortices are created by DC-induced electroosmosis. The analysis of particle separation is based on the classical excluded-volume macrotransport approach, which allows to derive the average particle velocity and the axial dispersion coefficient from the solution of two stationary advection-diffusion problems defined onto the channel cross section. We find that lateral vortices can enhance the separation efficiency quantitatively, e.g., by reducing the separation time of a two-species mixture by a 50-fold factor compared to standard OC-HDC.

UCTransNet: Rethinking the Skip Connections in U-Net from a Channel-Wise Perspective with Transformer

Most recent semantic segmentation methods adopt a U-Net framework with an encoder-decoder architecture. It is still challenging for U-Net with a simple skip connection scheme to model the global multi-scale context: 1) Not each skip connection setting is effective due to the issue of incompatible feature sets of encoder and decoder stage, even some skip connection negatively influence the segmentation performance; 2) The original U-Net is worse than the one without any skip connection on some datasets. Based on our findings, we propose a new segmentation framework, named UCTransNet (with a proposed CTrans module in U-Net), from the channel perspective with attention mechanism. Specifically, the CTrans (Channel Transformer) module is an alternate of the U-Net skip connections, which consists of a sub-module to conduct the multi-scale Channel Cross fusion with Transformer (named CCT) and a sub-module Channel-wise Cross-Attention (named CCA) to guide the fused multi-scale channel-wise information to effectively connect to the decoder features for eliminating the ambiguity. Hence, the proposed connection consisting of the CCT and CCA is able to replace the original skip connection to solve the semantic gaps for an accurate automatic medical image segmentation. The experimental results suggest that our UCTransNet produces more precise segmentation performance and achieves consistent improvements over the state-of-the-art for semantic segmentation across different datasets and conventional architectures involving transformer or U-shaped framework. Code: https://github.com/McGregorWwww/UCTransNet.

Three-channel (de)multiplexer for dense wavelength division (de)multiplexing based on a photonic crystal AAH cavity

We propose a three-channel (de)multiplexer for dense wavelength division (de)multiplexing based on a photonic crystal Aubry–André–Harper cavity. The transmission performance of the designed structure was modeled and analyzed according to coupled mode theory. The parameters of the proposed device are calculated using a three-dimensional finite-difference time-domain method for verification. Numerical results show that the minimum insertion loss and linewidth at operating wavelengths of 1556.2 nm, 1555.4 nm, and 1554.6 nm are 0.47 dB and 0.2 nm, respectively. The channel cross talk is lower than − 16.23 d B . The (de)multiplexer has the advantages of low loss, small size [ 19.35 × 11.53 ( µ m 2 ) per channel], and easy cascading structure, and can realize three-channel dense wavelength division (de)multiplexing with a channel spacing of 0.8 nm and be used in highly integrated large-capacity optical fiber communication systems.

A Novel CovidDetNet Deep Learning Model for Effective COVID-19 Infection Detection Using Chest Radiograph Images

The suspected cases of COVID-19 must be detected quickly and accurately to avoid the transmission of COVID-19 on a large scale. Existing COVID-19 diagnostic tests are slow and take several hours to generate the required results. However, on the other hand, most X-rays or chest radiographs only take less than 15 min to complete. Therefore, we can utilize chest radiographs to create a solution for early and accurate COVID-19 detection and diagnosis to reduce COVID-19 patient treatment problems and save time. For this purpose, CovidDetNet is proposed, which comprises ten learnable layers that are nine convolutional layers and one fully-connected layer. The architecture uses two activation functions: the ReLu activation function and the Leaky Relu activation function and two normalization operations that are batch normalization and cross channel normalization, making it a novel COVID-19 detection model. It is a novel deep learning-based approach that automatically and reliably detects COVID-19 using chest radiograph images. Towards this, a fine-grained COVID-19 classification experiment is conducted to identify and classify chest radiograph images into normal, COVID-19 positive, and pneumonia. In addition, the performance of the proposed novel CovidDetNet deep learning model is evaluated on a standard COVID-19 Radiography Database. Moreover, we compared the performance of our approach with hybrid approaches in which we used deep learning models as feature extractors and support vector machines (SVM) as a classifier. Experimental results on the dataset showed the superiority of the proposed CovidDetNet model over the existing methods. The proposed CovidDetNet outperformed the baseline hybrid deep learning-based models by achieving a high accuracy of 98.40%.

AN IMPROVED BUTTERFLY-NET BASED DETECTION OF STONES IN KIDNEY IMAGES

Kidney stone disease is one of the major life threatening problem existing worldwide. It is important to recognize the true and exact location of kidney stone. The ultrasound images of kidney comprise speckle noise and are of low contrast which makes the identification of kidney abnormalities a challenging task. Thus, the doctors may find recognition of small stones difficult. Butterfly-net model is a low-complexity convolutional neural network with sparse cross channel connections and with pre-processing techniques for kidney stone detection. IntegratingSerious problems are well approximated with network complexity based on frequency bandwidth rather than input dimension using a butterfly-net with a fully connected neural network. The proposed Butterfly-net obviously embraces the arranged construction in the Butterfly plot as the stacked convolutional layers. In the event that the boundaries are hard-coded as that in the Butterfly plot, Butterfly-net on the whole registers the Fourier coefficients of the info signal with ensured mathematical exactness. Keywords:Butterfly-Net , Kidney Stone, Deep Learning.