Surface Representation Research Articles

Identification of Facial Emotions Using Reinforcement model under Deep Learning

This paper addresses the identification of facial emotions using a reinforcement model under deep learning. Close-to-perception ability presents a more exhaustive recommendation on human-machine interaction (HMI). Because of the Transfer Self-training (TST), and the Representation Reinforcement Network (RRN), this study offers an active FER arrangement. Two modules are considered for depiction support arranging such as Surface Representation Reinforcement (SurRR) and Semantic Representation Reinforcement (SemaRR). SurRR highlights are detracting component communication centers in feature maps and match face attributes in different facets. Worldwide face settings are semantically sent in channel and dimensional facets of a piece. RRN has a limit concerning involved origin when the edges and computational complication are considerably belittled. Our technique was tried on informational indexes from CK+, RaFD, FERPLUS, and RAFDB, and it was viewed as 100 percent, 98.62 percent, 89.64 percent, and 88.72 percent, individually. Also, the early application exploration shows the way that our strategy can be utilized in HMI.

Influence of DEM Resolution on the Hydrological Responses of a Terraced Catchment: An Exploratory Modelling Approach

Terracing is widely used as an effective soil and water conservation practice in sloped terrains. Physically based hydrological models are useful tools for understanding the hydrological response of terraced catchments. These models typically require a DEM as input data, whose resolution is likely to influence the model accuracy. The main objective of the present work was to investigate how DEM resolution affects the accuracy of terrain representations and consequently the performance of SWAT hydrological model in simulating streamflow for a terraced eucalypt-dominated catchment (Portugal). Catchment´s hydrological responses were analyzed based on two contrasting topographic scenarios: terraces and no terrace, to evaluate the influence of terraces. To this end, different SWAT models were set up using multi-resolution DEMs (10 m, 1 m, 0.5 m, 0.25 m and 0.10 m) based on photogrammetric techniques and LiDAR data. LiDAR-derived DEMs (terraces scenario) improved topographic surface and watershed representation, consequently increasing the model performance, stage hydrographs and flow-duration curves accuracy. When comparing the contrasting topographic scenarios, SWAT simulations without terraces (10 m and 1 m DEMs) produced a more dynamic and rapid hydrological response. In this scenario, the streamflow was 28% to 36% higher than SWAT simulations accounting for the terraces, which corroborates the effectiveness of terraces as a water conservation practice.

3D MODELING OF GRASSLANDS THROUGH GNSS TECHNOLOGY

Monitoring and 3D modeling of an objective can only be done on the basis of high-precision topo-geodetic measurements, but also on the basis of UAV (drone) technologies, thus resulting in an overview of a certain study area. In this context, the present work aims to develop the creation of the 3D model of a grassland surface following a topographic survey by means of GNSS technology and specialized programs. This topographic survey was carried out with the Leica Viva GS07 GNSS equipment, an instrument that contains a wide range of applications. In this work, I used the Real Time Kinematic method to determine the points based on which I created the 3D model in AutoCAD using the TopoLT program. Because the data obtained is complex, in addition to 2D plans, we also have information that can be analyzed in 3D. For studies regarding the shape and the way of exploitation of the land, we created a digital model of the land which is the 3D digital representation of the respective surface. Based on the digital terrain model, we can easily visualize the relief of the grassland, in the present case the minimum altitude is 572.50 m, and the maximum altitude is 590.00 m, with a level difference of 17.5 m over the entire area. This digital terrain model can also be used to model water flow or other geomorphological processes that occur in the analyzed grassland (landslides, erosion, etc.). Multiple longitudinal or transversal profiles can be created over the entire surface to better observe the relief of the land.

The explicit form of the switching surface in admissible synthesis problem

In this article we consider the problem related to positional synthesis and controllability function method and more precisely to admissible maximum principle. Unlike the more common approach the admissible maximum principle method gives discontinuous solutions to the positional synthesis problem. Let us consider the canonical system of linear equations $\dot{x}_i=x_{i+1}, i=\overline{1,n-1}, \dot{x}_n=u$ with constraints $|u| \le d$. The problem for an arbitrary linear system $\dot{x} = A x + b u$ can be simplified to this problem for the canonical system. A controllability function $\Theta(x)$ is given as a unique positive solution of some equation $\Phi(x,\Theta) = 0$. The control is chosen to minimize derivative of the function $\Theta(x)$ and can be written as $u(x) = -d \text{ sign}(s(x,\Theta(x)))$. The set of points $s(x,\Theta(x)) = 0$ is called the switching surface, and it determines the points where control changes its sign. Normally it \mbox{contains} the variable $\Theta$ which is given implicitly as the solution of equation $\Phi(x, \Theta) = 0$. Our aim in this paper is to find a representation of the switching surface that does not depend on the function $\Theta(x)$. We call this representation the explicit form. In our case the expressions $\Phi(x, \Theta)$ and $s(x, \Theta)$ are both polynomials with respect to $\Theta$, so this problem is related to the problem of finding conditions when two polynomials have a common positive root. Earlier the solution for the 2-dimensional case was known. But during the exploration it was found out that for systems of higher dimensions there exist certain difficulties. In this article the switching surface for the three dimensional case is presented and researched. It is shown that this switching surface is a sliding surface (according to Filippov's definition). Also the other ways of constructing the switching surface using the interpolation and approximation are proposed and used for finding the trajectories of concrete points.

FLUID FLOW MODELLING WITH FREE SURFACE

Fluid is a substance that can flow in the form of a liquid or a gas. Based on the movement of the fluid is divided into static and dynamic fluids. This study discusses fluid dynamics, namely modelling fluid flow accompanied by a free surface and an obstacle in the fluid flow. Fluid modelling generally makes some basic assumptions into mathematical equations. The assumptions are incompressible, steady-state and irrotational. The steps to obtain a fluid flow model are using Newton’s second law, the law of conservation of mass, and the law of conservation of momentum to obtain the general Navier-Stokes equation, the designing the Euler free surface equation, the Bernoulli equation, then making a free surface representation and linearizing the wave equation so that it is obtained fluid flow model. The resulting mathematical model is a Laplace equation with boundary conditions in the fluid.

Reliable emulation of complex functionals by active learning with error control.

A statistical emulator can be used as a surrogate of complex physics-based calculations to drastically reduce the computational cost. Its successful implementation hinges on an accurate representation of the nonlinear response surface with a high-dimensional input space. Conventional "space-filling" designs, including random sampling and Latin hypercube sampling, become inefficient as the dimensionality of the input variables increases, and the predictive accuracy of the emulator can degrade substantially for a test input distant from the training input set. To address this fundamental challenge, we develop a reliable emulator for predicting complex functionals by active learning with error control (ALEC). The algorithm is applicable to infinite-dimensional mapping with high-fidelity predictions and a controlled predictive error. The computational efficiency has been demonstrated by emulating the classical density functional theory (cDFT) calculations, a statistical-mechanical method widely used in modeling the equilibrium properties of complex molecular systems. We show that ALEC is much more accurate than conventional emulators based on the Gaussian processes with "space-filling" designs and alternative active learning methods. In addition, it is computationally more efficient than direct cDFT calculations. ALEC can be a reliable building block for emulating expensive functionals owing to its minimal computational cost, controllable predictive error, and fully automatic features.

Remeshing flexible membranes under the control of free energy.

Cell membranes are flexible and often undergo large-scale morphological changes during processes like mitosis, protrusion and retraction, or vesicle fusion. Mathematical modeling of cell membranes depends on a representation of the free-form surface by discrete meshes. During morphological changes, these meshes must be adjusted under the minimization of the total free energy. Current methodology for meshing is limited in one of two ways: 1) Free energy-dependent methods have no restriction on the mesh geometry. The resulting irregular meshes cause artifacts in follow-up models of morphodynamics. 2) Geometry-dependent methods maintain mesh quality but violate the physics of free energy minimization. To fill this gap, we regulate mesh geometries via a free-energy-determined remeshing process: adding and removing mesh elements upon morphological changes based on barrier crossings in a double-barrier potential between neighboring vertices in the meshes. We test the method's robustness by reproducing the morphodynamics of red blood cells and vesicle fusions; and we demonstrate the method's adaptability by simulating the formation of filopodia, lamellipodia and invaginations. Finally, we use the method to study a mechanical decoupling effect of two connected membrane tethers that has been recently observed experimentally, but has not been mechanistically explained in the context of a complete membrane surface. We propose a biophysical model that strengthens the decoupling effect and broadens the original interpretation of the experiment. The method is developed in C/Matlab and distributed via https://github.com/DanuserLab/biophysicsModels.

Explainable Deep Reinforcement Learning: State of the Art and Challenges

Interpretability, explainability, and transparency are key issues to introducing artificial intelligence methods in many critical domains. This is important due to ethical concerns and trust issues strongly connected to reliability, robustness, auditability, and fairness, and has important consequences toward keeping the human in the loop in high levels of automation, especially in critical cases for decision making, where both (human and the machine) play important roles. Although the research community has given much attention to explainability of closed (or black) prediction boxes, there are tremendous needs for explainability of closed-box methods that support agents to act autonomously in the real world. Reinforcement learning methods, and especially their deep versions, are such closed-box methods. In this article, we aim to provide a review of state-of-the-art methods for explainable deep reinforcement learning methods, taking also into account the needs of human operators—that is, of those who make the actual and critical decisions in solving real-world problems. We provide a formal specification of the deep reinforcement learning explainability problems, and we identify the necessary components of a general explainable reinforcement learning framework. Based on these, we provide a comprehensive review of state-of-the-art methods, categorizing them into classes according to the paradigm they follow, the interpretable models they use, and the surface representation of explanations provided. The article concludes by identifying open questions and important challenges.

Experimental Case Study of Stochastic Surrogate-Assisted Engine Calibration

Reducing experimental calibration cost to achieve optimal system performance is challenging, especially for highly nonlinear engine systems. With the advancement and widespread adoption of machine learning methods for control applications, it is now possible to use an intelligent black-box model to efficiently optimize nonlinear control systems without the detailed knowledge of system dynamics. The surrogate-assisted optimization approach has been recently proven to work effectively to reduce the overall experimental budget (cost). However, its application to calibrate and optimize a practical control system experimentally is a challenge due to stochastic system and measurement noises. This article experimentally calibrates and optimizes the control parameters of a 6.7L Ford diesel engine using the stochastic surrogate-assisted optimization automatically based on a stochastic Kriging model, which is the first step toward automating the engine calibration process. Two engine control variables, namely, exhaust gas recirculation valve and variable geometry turbocharger vane positions, are calibrated to obtain a tradeoff between engine efficiency, in terms of brake-specific fuel consumption, and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\text{NO}_{x}$</tex-math></inline-formula> emissions, utilizing both the deterministic and stochastic surrogate-assisted learning algorithms. Experimental results show a much better representation of the output response surfaces with the stochastic surrogate-assisted optimization than that from the deterministic approach. Note that the stochastic approach helps direct the search toward the global optimal region without wasting the evaluation budget in exploring the local optima, and the deterministic approach cannot effectively handle stochastic system and measurement noises, resulting in overfitting the data. Therefore, to solve a practical optimization problem with noises, stochastic surrogate-assisted optimization is recommended.

Towards monitoring the CO2 source–sink distribution over India via inverse modelling: quantifying the fine-scale spatiotemporal variability in the atmospheric CO2 mole fraction

Abstract. Improving the estimates of CO2 sources and sinks over India through inverse methods calls for a comprehensive atmospheric monitoring system involving atmospheric transport models that make a realistic accounting of atmospheric CO2 variability along with a good coverage of ground-based monitoring stations. This study investigates the importance of representing fine-scale variability in atmospheric CO2 in models for the optimal use of observations through inverse modelling. The unresolved variability in atmospheric CO2 in coarse models is quantified by using WRF-Chem (Weather Research and Forecasting model coupled with Chemistry) simulations at a spatial resolution of 10 km × 10 km. We show that the representation errors due to unresolved variability in the coarse model with a horizontal resolution of 1∘ (∼ 100 km) are considerable (median values of 1.5 and 0.4 ppm, parts per million, for the surface and column CO2, respectively) compared to the measurement errors. The monthly averaged surface representation error reaches up to ∼ 5 ppm, which is even comparable to half of the magnitude of the seasonal variability or concentration enhancement due to hotspot emissions. Representation error shows a strong dependence on multiple factors such as time of the day, season, terrain heterogeneity, and changes in meteorology and surface fluxes. By employing a first-order inverse modelling scheme using pseudo-observations from nine tall-tower sites over India, we show that the net ecosystem exchange (NEE) flux uncertainty solely due to unresolved variability is in the range of 3.1 % to 10.3 % of the total NEE of the region. By estimating the representation error and its impact on flux estimations during different seasons, we emphasize the need to take account of fine-scale CO2 variability in models over the Indian subcontinent to better understand processes regulating CO2 sources and sinks. The efficacy of a simple parameterization scheme is further demonstrated to capture these unresolved variations in coarse models.

Position-Based Surface Tension Flow

This paper presents a novel approach to simulating surface tension flow within a position-based dynamics (PBD) framework. We enhance the conventional PBD fluid method in terms of its surface representation and constraint enforcement to furnish support for the simulation of interfacial phenomena driven by strong surface tension and contact dynamics. The key component of our framework is an on-the-fly local meshing algorithm to build the local geometry around each surface particle. Based on this local mesh structure, we devise novel surface constraints that can be integrated seamlessly into a PBD framework to model strong surface tension effects. We demonstrate the efficacy of our approach by simulating a multitude of surface tension flow examples exhibiting intricate interfacial dynamics of films and drops, which were all infeasible for a traditional PBD method.

A Fully Coupled Thermomechanical Phase Field Method for Modeling Cracks with Frictional Contact

In this paper, a thermomechanical coupled phase field method is developed to model cracks with frictional contact. Compared to discrete methods, the phase field method can represent arbitrary crack geometry without an explicit representation of the crack surface. The two distinguishable features of the proposed phase field method are: (1) for the mechanical phase, no specific algorithm is needed for imposing contact constraints on the fracture surfaces; (2) for the thermal phase, formulations are proposed for incorporating the phase field damage parameter so that different thermal conductance conditions are accommodated. While the stress is updated explicitly in the regularized interface regions under different contact conditions, the thermal conductivity is determined under different conductance conditions. In particular, we consider a pressure-dependent thermal conductance model (PDM) that is fully coupled with the mechanical phase, along with the other three thermal conductance models, i.e., the fully conductive model (FCM), the adiabatic model (ACM), and the uncoupled model (UCM). The potential of this formulation is showcased by several benchmark problems. We gain insights into the role of the temperature field affecting the mechanical field. Several 2D boundary value problems are addressed, demonstrating the model’s ability to capture cracking phenomena with the effect of the thermal field. We compare our results with the discrete methods as well as other phase field methods, and a very good agreement is achieved.

Soil liquefaction potential assessment using ambient noise: A case study in Pohang, Korea

An MW 5.5 earthquake occurred in Pohang, Korea, on 15 November 2017 and caused large damage. In and around the epicentral area the earthquake also remained numerous sand boils, which is a surface representation of soil liquefaction at shallow depth during the earthquake. Soil liquefaction is one of the most dangerous consequences of an earthquake. Here we show that the spatial distribution of the ground vulnerability index (Kg), which we estimate via ambient noise analysis, correlates very well with the distribution of sand boils. Our Kg model based on the dense microtremor surveys at differing geological conditions and urbanizations indicates that only 28.4% of the study area is vulnerable to ground shaking (Kg &amp;gt; 20), with 91.1% of the observed sand boils occurring in these vulnerable areas. We also observed that Kg values estimated in the study correlate well with both VS30 values and geologic units in the study area. This case report confirmed that the Kg can be an affordable supplement to traditional, but expensive and time-consuming, geotechnical/geophysical techniques for the initial screening and regional evaluations. Such a Kg map can assist stakeholders in earthquake-prone regions in identifying areas more susceptible to liquefaction and bring a recommendation to consider seismic hazard mitigation.

Graph-Theoretic Molecular Fragmentation for Potential Surfaces Leads Naturally to a Tensor Network Form and Allows Accurate and Efficient Quantum Nuclear Dynamics.

Molecular fragmentation methods have revolutionized quantum chemistry. Here, we use a graph-theoretically generated molecular fragmentation method, to obtain accurate and efficient representations for multidimensional potential energy surfaces and the quantum time-evolution operator, which plays a critical role in quantum chemical dynamics. In doing so, we find that the graph-theoretic fragmentation approach naturally reduces the potential portion of the time-evolution operator into a tensor network that contains a stream of coupled lower-dimensional propagation steps to potentially achieve quantum dynamics with reduced complexity. Furthermore, the fragmentation approach used here has previously been shown to allow accurate and efficient computation of post-Hartree-Fock electronic potential energy surfaces, which in many cases has been shown to be at density functional theory cost. Thus, by combining the advantages of molecular fragmentation with the tensor network formalism, the approach yields an on-the-fly quantum dynamics scheme where both the electronic potential calculation and nuclear propagation portion are enormously simplified through a single stroke. The method is demonstrated by computing approximations to the propagator and to potential surfaces for a set of coupled nuclear dimensions within a protonated water wire problem exhibiting the Grotthuss mechanism of proton transport. In all cases, our approach has been shown to reduce the complexity of representing the quantum propagator, and by extension action of the propagator on an initial wavepacket, by several orders, with minimal loss in accuracy.

A DFT Characterization of Structural, Mechanical, and Thermodynamic Properties of Ag9In4 Binary Intermetallic Compound

The intermetallic compounds (IMCs) at the interface between the solder joint and metal bond pad/under bump metallization (UBM) exert a significant impact on the thermal–mechanical behavior of microelectronic packages because of their unique physical properties. In this study, a theoretical investigation of the physical properties, namely structural, mechanical, and thermodynamic properties, of the Ag9In4 IMC was conducted using ab initio density functional theory (DFT) calculations. The calculated equilibrium lattice constants were in good agreement with the literature experimental data. Furthermore, with the calculated elastic constants, we can derive the ductility and brittleness nature, elastic anisotropy, and direction-dependent elastic properties of Ag9In4 through several elastic indices, three-dimensional surface representation, and two-dimensional projections of elastic properties. The calculations inferred that the cubic Ag9In4 IMC confers structural and mechanical stability, ductility, relative low stiffness and hardness, and elastic anisotropy. Finally, the thermodynamic properties, i.e., Debye temperature, heat capacity, and minimum thermal conductivity, were also investigated. Evidently, the low-temperature heat capacity conforms to the Debye heat capacity theory and the high-temperature one complies with the classical Dulong–Petit law.

Comparative Analysis of Uncoupled Mode Approximations for Molecular Thermochemistry and Kinetics.

The accurate prediction of thermochemistry and kinetic parameters is an important task for reaction modeling. Unfortunately, the commonly used harmonic oscillator model is often not accurate enough due to the absence of anharmonic effects. In this work, we improve the representation of an anharmonic potential energy surface (PES) using uncoupled mode (UM) approximations, which model the full-dimensional PES as a sum of one-dimensional potentials of each mode. We systematically analyze different PES sampling schemes and coordinate systems for constructing the one-dimensional potentials, and benchmark the performance of UM methods on data sets of molecular thermochemistry and kinetic properties. The results show that the accuracy of the UM approach strongly depends on how the one-dimensional potentials are defined. If one-dimensional potentials are constructed by sampling along normal mode directions (UM-N) or along the directions that minimize intermode coupling (E- and E'-optimized), the accuracies of the predicted properties are not significantly improved compared to the harmonic oscillator model. However, significant improvements can be achieved by sampling the torsional modes separately from the vibrational modes (UM-T and UM-VT). In this work, three types of coordinate systems are examined, including redundant internal coordinates (RIC), hybrid internal coordinates (HIC), and translation-rotation-internal coordinates (TRIC). The HIC and TRIC coordinate systems can outperform RIC since transition state species may contain large-amplitude interfragmentary motions that regular internal coordinates can not describe adequately. Among all the methods we examined, the activation energies and pre-exponential factors calculated using UM-VT with either TRIC or HIC best agree with the reference values. Since UM-VT requires only a number of additional single point energy calculations for each independent mode, the scaling of computational costs of UM-VT is the same as that of the standard harmonic oscillator model, making UM-VT an appealing way of calculating the thermochemistry and kinetic properties for large-size systems.

Concrete Bridge Crack Image Classification Using Histograms of Oriented Gradients, Uniform Local Binary Patterns, and Kernel Principal Component Analysis

Bridges deteriorate over time, which requires the continuous monitoring of their condition. There are many digital technologies for inspecting and monitoring bridges in real-time. In this context, computer vision has extensively studied cracks to automate their identification in concrete surfaces, overcoming the conventional manual methods that rely on human judgment. The general framework of vision-based techniques consists of feature extraction using different filters and descriptors and classifier training to perform the classification task. However, training can be time-consuming and computationally expensive, depending on the dimension of the features. To address this limitation, dimensionality reduction techniques are applied to extracted features, and a new feature subspace is generated. This work used histograms of oriented gradients (HOGs) and uniform local binary patterns (ULBPs) to extract features from a dataset containing over 3000 uncracked and cracked images covering different patterns of cracks and concrete surface representations. Nonlinear dimensionality reduction was performed using kernel principal component analysis (KPCA), and three machine learning classifiers were implemented to conduct the classification. The experimental results show that the classification scheme based on the support-vector machine (SVM) model and feature-level fusion of the HOG and ULBP features after KPCA application provided the best results as an accuracy of 99.26% was achieved by the proposed classification framework.

A topological solution to object segmentation and tracking

The world is composed of objects, the ground, and the sky. Visual perception of objects requires solving two fundamental challenges: 1) segmenting visual input into discrete units and 2) tracking identities of these units despite appearance changes due to object deformation, changing perspective, and dynamic occlusion. Current computer vision approaches to segmentation and tracking that approach human performance all require learning, raising the question, Can objects be segmented and tracked without learning? Here, we show that the mathematical structure of light rays reflected from environment surfaces yields a natural representation of persistent surfaces, and this surface representation provides a solution to both the segmentation and tracking problems. We describe how to generate this surface representation from continuous visual input and demonstrate that our approach can segment and invariantly track objects in cluttered synthetic video despite severe appearance changes, without requiring learning.

SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces From RGB Images.

This paper focuses on the challenging task of learning 3D object surface reconstructions from RGB images. Existing methods achieve varying degrees of success by using different surface representations. However, they all have their own drawbacks, and cannot properly reconstruct the surface shapes of complex topologies, arguably due to a lack of constraints on the topological structures in their learning frameworks. To this end, we propose to learn and use the topology-preserved, skeletal shape representation to assist the downstream task of object surface reconstruction from RGB images. Technically, we propose the novel SkeletonNet design that learns a volumetric representation of a skeleton via a bridged learning of a skeletal point set, where we use parallel decoders each responsible for the learning of points on 1D skeletal curves and 2D skeletal sheets, as well as an efficient module of globally guided subvolume synthesis for a refined, high-resolution skeletal volume; we present a differentiable Point2Voxel layer to make SkeletonNet end-to-end and trainable. With the learned skeletal volumes, we propose two models, the Skeleton-Based Graph Convolutional Neural Network (SkeGCNN) and the Skeleton-Regularized Deep Implicit Surface Network (SkeDISN), which respectively build upon and improve over the existing frameworks of explicit mesh deformation and implicit field learning for the downstream surface reconstruction task. We conduct thorough experiments that verify the efficacy of our proposed SkeletonNet. SkeGCNN and SkeDISN outperform existing methods as well, and they have their own merits when measured by different metrics. Additional results in generalized task settings further demonstrate the usefulness of our proposed methods. We have made our implementation code publicly available at https://github.com/tangjiapeng/SkeletonNet.

Analysis based on neural representation of natural object surfaces to elucidate the mechanisms of a trained AlexNet model.

Analysis and understanding of trained deep neural networks (DNNs) can deepen our understanding of the visual mechanisms involved in primate visual perception. However, due to the limited availability of neural activity data recorded from various cortical areas, the correspondence between the characteristics of artificial and biological neural responses for visually recognizing objects remains unclear at the layer level of DNNs. In the current study, we investigated the relationships between the artificial representations in each layer of a trained AlexNet model (based on a DNN) for object classification and the neural representations in various levels of visual cortices such as the primary visual (V1), intermediate visual (V4), and inferior temporal cortices. Furthermore, we analyzed the profiles of the artificial representations at a single channel level for each layer of the AlexNet model. We found that the artificial representations in the lower-level layers of the trained AlexNet model were strongly correlated with the neural representation in V1, whereas the responses of model neurons in layers at the intermediate and higher-intermediate levels of the trained object classification model exhibited characteristics similar to those of neural activity in V4 neurons. These results suggest that the trained AlexNet model may gradually establish artificial representations for object classification through the hierarchy of its network, in a similar manner to the neural mechanisms by which afferent transmission beginning in the low-level features gradually establishes object recognition as signals progress through the hierarchy of the ventral visual pathway.