Search results for: Deep Neural Network

Authors: Riznaldi Akbar

Abstract:

In this study, we compared the performance of the Artificial Neural Network (ANN) model with back-propagation algorithm in correctly predicting in-sample and out-of-sample external debt crisis in Indonesia. We found that exchange rate, foreign reserves, and exports are the major determinants to experiencing external debt crisis. The ANN in-sample performance provides relatively superior results. The ANN model is able to classify correctly crisis of 89.12 per cent with reasonably low false alarms of 7.01 per cent. In out-of-sample, the prediction performance fairly deteriorates compared to their in-sample performances. It could be explained as the ANN model tends to over-fit the data in the in-sample, but it could not fit the out-of-sample very well. The 10-fold cross-validation has been used to improve the out-of-sample prediction accuracy. The results also offer policy implications. The out-of-sample performance could be very sensitive to the size of the samples, as it could yield a higher total misclassification error and lower prediction accuracy. The ANN model could be used to identify past crisis episodes with some accuracy, but predicting crisis outside the estimation sample is much more challenging because of the presence of uncertainty.

Keywords: debt crisis, external debt, artificial neural network, ANN

Procedia PDF Downloads 423

Authors: Javier Fernandez De Canete, Alvaro Fernandez-Quintero

Abstract:

Nowadays, adaptive control schemes are being used when model based control schemes are applied in presence of uncertainty and model mismatches. Artificial neural networks have been employed both in modelling and control of non-linear dynamic systems with unknown dynamics. In fact, these are powerful tools to solve this control problem when only input-output operational data are available. A neural network controller under SIMULINK together with the ARDUINO hardware platform has been used to perform real-time speed control of a computer case fan. Comparison of performance with a PID controller has also been presented in order to show the efficacy of neural control under different command signals tracking and also when disturbance signals are present in the speed control loops.

Keywords: neural networks, ARDUINO platform, SIMULINK, adaptive speed control

Procedia PDF Downloads 335

Authors: Sezin Ekşioğlu

Abstract:

Multilabel classification has a huge importance for several applications, it is also a challenging research topic. It is a kind of supervised learning that contains binary targets. The distance between multilabel and binary classification is having more than one class in multilabel classification problems. Features can belong to one class or many classes. There exists a wide range of applications for multi label prediction such as image labeling, text categorization, gene functionality. Even though features are classified in many classes, they may not always be properly classified. There are many ensemble methods for the classification. However, most of the researchers have been concerned about better multilabel methods. Especially little ones focus on both efficiency of classifiers and pairwise relationships at the same time in order to implement better multilabel classification. In this paper, we worked on modified ensemble methods by getting benefit from k-Nearest Neighbors and neural network structure to address issues within a beneficial way and to get better impacts from the multilabel classification. Publicly available datasets (yeast, emotion, scene and birds) are performed to demonstrate the developed algorithm efficiency and the technique is measured by accuracy, F1 score and hamming loss metrics. Our algorithm boosts benchmarks for each datasets with different metrics.

Keywords: multilabel, classification, neural network, KNN

Procedia PDF Downloads 134

Authors: Lea Lenhardt, Ivana Zeković, Tatjana Dramićanin, Miroslav D. Dramićanin

Abstract:

Fluorescence spectroscopy coupled with parallel factor analysis (PARAFAC) and artificial neural networks (ANN) were used for characterization and classification of honey. Excitation emission spectra were obtained for 95 honey samples of different botanical origin (acacia, sunflower, linden, meadow, and fake honey) by recording emission from 270 to 640 nm with excitation in the range of 240-500 nm. Fluorescence spectra were described with a six-component PARAFAC model, and PARAFAC scores were further processed with two types of ANN’s (feed-forward network and self-organizing maps) to obtain algorithms for classification of honey on the basis of their botanical origin. Both ANN’s detected fake honey samples with 100% sensitivity and specificity.

Keywords: honey, fluorescence, PARAFAC, artificial neural networks

Procedia PDF Downloads 933

Authors: Randy Price, Harbir Antil, Rainald Löhner, Fumiya Togashi

Abstract:

Chemically reacting flows are common in engineering applications such as hypersonic flow, combustion, explosions, manufacturing process, and environmental assessments. The number of reactions in combustion simulations can exceed 100, making a large number of flow and combustion problems beyond the capabilities of current supercomputers. Motivated by this, deep neural networks (DNNs) will be introduced with the goal of eventually replacing the existing chemistry software packages with DNNs. The DNNs used in this paper are motivated by the Residual Neural Network (ResNet) architecture. In the continuum limit, ResNets become an optimization problem constrained by an ODE. Such a feature allows the use of ODE control techniques to enhance the DNNs. In this work, DNNs are constructed, which update the species un at the nᵗʰ timestep to uⁿ⁺¹ at the n+1ᵗʰ timestep. Parallel DNNs are trained for each species, taking in uⁿ as input and outputting one component of uⁿ⁺¹. These DNNs are applied to multiple species and reactions common in chemically reacting flows such as H₂-O₂ reactions. Experimental results show that the DNNs are able to accurately replicate the dynamics in various situations and in the presence of errors.

Keywords: chemical reacting flows, computational fluid dynamics, ODEs, residual neural networks, ResNets

Procedia PDF Downloads 100

Authors: F. Massri, M. Shkarneh, B. Almassri

Abstract:

Water scarcity and increasing water demand especially for residential use are major challenges facing Palestine. The need to accurately forecast water consumption is useful for the planning and management of this natural resource. The main objective of this paper is to (i) study the major factors influencing the water consumption in Palestine, (ii) understand the general pattern of Household water consumption, (iii) assess the possible changes in household water consumption and suggest appropriate remedies and (iv) develop prediction model based on the Artificial Neural Network to the water consumption in Palestinian cities. The paper is organized in four parts. The first part includes literature review of household water consumption studies. The second part concerns data collection methodology, conceptual frame work for the household water consumption surveys, survey descriptions and data processing methods. The third part presents descriptive statistics, multiple regression and analysis of the water consumption in the two Palestinian cities. The final part develops the use of Artificial Neural Network for modeling the water consumption in Palestinian cities.

Keywords: water management, demand forecasting, consumption, ANN, Ramallah

Procedia PDF Downloads 194

Authors: Mehdi Montakhabrazlighi, Ercan Balikci

Abstract:

The neural network (NN) method is applied to alloy development of single crystal Ni-base Superalloys with low density and improved mechanical strength. A set of 1200 dataset which includes chemical composition of the alloys, applied stress and temperature as inputs and density and time to rupture as outputs is used for training and testing the network. Thermodynamic phase diagram modeling of the screened alloys is performed with Thermocalc software to model the equilibrium phases and also microsegregation in solidification processing. The model is first trained by 80% of the data and the 20% rest is used to test it. Comparing the predicted values and the experimental ones showed that a well-trained network is capable of accurately predicting the density and time to rupture strength of the Ni-base superalloys. Modeling results is used to determine the effect of alloying elements, stress, temperature and gamma-prime phase volume fraction on rupture strength of the Ni-base superalloys. This approach is in line with the materials genome initiative and integrated computed materials engineering approaches promoted recently with the aim of reducing the cost and time for development of new alloys for critical aerospace components. This work has been funded by TUBITAK under grant number 112M783.

Keywords: neural network, rupture strength, superalloy, thermocalc

Procedia PDF Downloads 289

Authors: Souvik Phadikar, Nidul Sinha, Rajdeep Ghosh

Abstract:

In brainwaves research using electroencephalogram (EEG) signals, finding the most relevant and effective feature set for identification of activities in the human brain is a big challenge till today because of the random nature of the signals. The feature extraction method is a key issue to solve this problem. Finding those features that prove to give distinctive pictures for different activities and similar for the same activities is very difficult, especially for the number of activities. The performance of a classifier accuracy depends on this quality of feature set. Further, more number of features result in high computational complexity and less number of features compromise with the lower performance. In this paper, a novel idea of the selection of optimal feature set using a heuristically optimized deep autoencoder is presented. Using various feature extraction methods, a vast number of features are extracted from the EEG signals and fed to the autoencoder deep neural network. The autoencoder encodes the input features into a small set of codes. To avoid the gradient vanish problem and normalization of the dataset, a meta-heuristic search algorithm is used to minimize the mean square error (MSE) between encoder input and decoder output. To reduce the feature set into a smaller one, 4 hidden layers are considered in the autoencoder network; hence it is called Heuristically Optimized Deep Autoencoder (HO-DAE). In this method, no features are rejected; all the features are combined into the response of responses of the hidden layer. The results reveal that higher accuracy can be achieved using optimal reduced features. The proposed HO-DAE is also compared with the regular autoencoder to test the performance of both. The performance of the proposed method is validated and compared with the other two methods recently reported in the literature, which reveals that the proposed method is far better than the other two methods in terms of classification accuracy.

Keywords: autoencoder, brainwave signal analysis, electroencephalogram, feature extraction, feature selection, optimization

Procedia PDF Downloads 97

Authors: Giselle Maggie-Fer Castañeda Lozano

Abstract:

The main objective of this research was to explore the role of neural network architectures in simulating behavioral phenomena as a potential explanation for selective associations, specifically related to biological constraints on learning. Biological constraints on learning refer to the limitations observed in conditioning procedures, where learning is expected to occur. The study involved simulations of five different experiments exploring various phenomena and sources of biological constraints in learning. These simulations included the interaction between response and reinforcer, stimulus and reinforcer, specificity of stimulus-reinforcer associations, species differences, neuroanatomical constraints, and learning in uncontrolled conditions. The overall results demonstrated that by manipulating neural network architectures, conditions can be created to model and explain diverse biological constraints frequently reported in comparative psychology literature as learning typicities. Additionally, the simulations offer predictive content worthy of experimental testing in the pursuit of new discoveries regarding the specificity of learning. The implications and limitations of these findings are discussed. Finally, it is suggested that this research could inaugurate a line of inquiry involving the use of neural networks to study biological factors in behavior, fostering the development of more ethical and precise research practices.

Keywords: comparative psychology, connectionism, conditioning, experimental analysis of behavior, neural networks

Procedia PDF Downloads 44

Authors: Yash Jain

Abstract:

The US Center for Disease Control has recommended wearing masks to slow the spread of the virus. The research uses a video feed from a camera to conduct real-time classifications of whether or not a human is correctly wearing a mask, incorrectly wearing a mask, or not wearing a mask at all. Utilizing two distinct datasets from the open-source website Kaggle, a mask detection network had been trained. The first dataset that was used to train the model was titled 'Face Mask Detection' on Kaggle, where the dataset was retrieved from and the second dataset was titled 'Face Mask Dataset, which provided the data in a (YOLO Format)' so that the TinyYoloV3 model could be trained. Based on the data from Kaggle, two machine learning models were implemented and trained: a Tiny YoloV3 Real-time model and a two-stage neural network classifier. The two-stage neural network classifier had a first step of identifying distinct faces within the image, and the second step was a classifier to detect the state of the mask on the face and whether it was worn correctly, incorrectly, or no mask at all. The TinyYoloV3 was used for the live feed as well as for a comparison standpoint against the previous two-stage classifier and was trained using the darknet neural network framework. The two-stage classifier attained a mean average precision (MAP) of 80%, while the model trained using TinyYoloV3 real-time detection had a mean average precision (MAP) of 59%. Overall, both models were able to correctly classify stages/scenarios of no mask, mask, and incorrectly worn masks.

Keywords: datasets, classifier, mask-detection, real-time, TinyYoloV3, two-stage neural network classifier

Procedia PDF Downloads 134

Authors: Abhisek Dash

Abstract:

This study had three neural networks, one for number segmentation, one for number detection and one for number recognition all of which are coupled to one another. All networks were trained on the MNIST dataset and were convolutional. It was assumed that the images had lighter background and darker foreground. The segmentation network took 28x28 images as input and had sixteen outputs. Segmentation training starts when a dark pixel is encountered. Taking a window(7x7) over that pixel as focus, the eight neighborhood of the focus was checked for further dark pixels. The segmentation network was then trained to move in those directions which had dark pixels. To this end the segmentation network had 16 outputs. They were arranged as “go east”, ”don’t go east ”, “go south east”, “don’t go south east”, “go south”, “don’t go south” and so on w.r.t focus window. The focus window was resized into a 28x28 image and the network was trained to consider those neighborhoods which had dark pixels. The neighborhoods which had dark pixels were pushed into a queue in a particular order. The neighborhoods were then popped one at a time stitched to the existing partial image of the number one at a time and trained on which neighborhoods to consider when the new partial image was presented. The above process was repeated until the image was fully covered by the 7x7 neighborhoods and there were no more uncovered black pixels. During testing the network scans and looks for the first dark pixel. From here on the network predicts which neighborhoods to consider and segments the image. After this step the group of neighborhoods are passed into the detection network. The detection network took 28x28 images as input and had two outputs denoting whether a number was detected or not. Since the ground truth of the bounds of a number was known during training the detection network outputted in favor of number not found until the bounds were not met and vice versa. The recognition network was a standard CNN that also took 28x28 images and had 10 outputs for recognition of numbers from 0 to 9. This network was activated only when the detection network votes in favor of number detected. The above methodology could segment connected and overlapping numbers. Additionally the recognition unit was only invoked when a number was detected which minimized false positives. It also eliminated the need for rules of thumb as segmentation is learned. The strategy can also be extended to other characters as well.

Keywords: convolutional neural networks, OCR, text detection, text segmentation

Procedia PDF Downloads 138

Authors: Gulfam Haider, sana danish

Abstract:

Convolutional Neural Networks (CNNs) have emerged as powerful tools for image classification, and the choice of optimizers profoundly affects their performance. The study of optimizers and their adaptations remains a topic of significant importance in machine learning research. While numerous studies have explored and advocated for various optimizers, the efficacy of these optimization techniques is still subject to scrutiny. This work aims to address the challenges surrounding the effectiveness of optimizers by conducting a comprehensive analysis and evaluation. The primary focus of this investigation lies in examining the performance of different optimizers when employed in conjunction with the popular activation function, Rectified Linear Unit (ReLU). By incorporating ReLU, known for its favorable properties in prior research, the aim is to bolster the effectiveness of the optimizers under scrutiny. Specifically, we evaluate the adjustment of these optimizers with both the original Softmax activation function and the modified ReLU activation function, carefully assessing their impact on overall performance. To achieve this, a series of experiments are conducted using a well-established benchmark dataset for image classification tasks, namely the Canadian Institute for Advanced Research dataset (CIFAR-10). The selected optimizers for investigation encompass a range of prominent algorithms, including Adam, Root Mean Squared Propagation (RMSprop), Adaptive Learning Rate Method (Adadelta), Adaptive Gradient Algorithm (Adagrad), and Stochastic Gradient Descent (SGD). The performance analysis encompasses a comprehensive evaluation of the classification accuracy, convergence speed, and robustness of the CNN models trained with each optimizer. Through rigorous experimentation and meticulous assessment, we discern the strengths and weaknesses of the different optimization techniques, providing valuable insights into their suitability for image classification tasks. By conducting this in-depth study, we contribute to the existing body of knowledge surrounding optimizers in CNNs, shedding light on their performance characteristics for image classification. The findings gleaned from this research serve to guide researchers and practitioners in making informed decisions when selecting optimizers and activation functions, thus advancing the state-of-the-art in the field of image classification with convolutional neural networks.

Keywords: deep neural network, optimizers, RMsprop, ReLU, stochastic gradient descent

Procedia PDF Downloads 97

Authors: Ge Zheng, Jun Peng

Abstract:

Unbalanced power output from individual cells of Proton Exchange Membrane Fuel Cell (PEMFC) has direct effects on PEMFC stack performance, in particular under transient operation. In the paper, a multi-layer ANN (Artificial Neural Network) model Radial Basis Functions (RBF) has been developed for predicting cells' output profiles by applying gas supply parameters, cooling conditions, temperature measurement of individual cells, etc. The feed-forward ANN model was validated with experimental data. Influence of relevant parameters of RBF on the network accuracy was investigated. After adequate model training, the modelling results show good correspondence between actual measurements and reconstructed output profiles. Finally, after the model was used to optimize the stack output performance under steady-state and transient operating conditions, it suggested that the developed ANN control model can help PEMFC stack to have obvious improvement on power output under fast acceleration process.

Keywords: proton exchange membrane fuel cell, PEMFC, artificial neural network, ANN, cell output profile, transient

Procedia PDF Downloads 150

Authors: Deba Kumar Sarma, Sanjib Kr. Rajbongshi

Abstract:

Surface finish is an important product quality in machining. At first, experiments were carried out to investigate the effect of the cutting tool nose radius (considering 1mm and 0.65mm) in prediction of surface finish with process parameters of cutting speed, feed and depth of cut. For all possible cutting conditions, full factorial design was considered as two levels four parameters. Commercial Mild Steel bar and High Speed Steel (HSS) material were considered as work-piece and cutting tool material respectively. In order to obtain functional relationship between process parameters and surface roughness, neural network was used which was found to be capable for the prediction of surface roughness within a reasonable degree of accuracy. It was observed that tool nose radius of 1mm provides better surface finish in comparison to 0.65 mm. Also, it was observed that feed rate has a significant influence on surface finish.

Keywords: full factorial design, neural network, nose radius, surface finish

Procedia PDF Downloads 348

Authors: Henni Mansour Abdelwaheb

Abstract:

This study discusses the design of a bounded position/force feedback controller developed to ensure position and force tracking for bilateral teleoperation arms operating with variable delay, and actuator saturation. Also, an adaptive robust Radial Basis Function (RBF) neural network is used to estimate the environment torque. The parameters of the environment torque are then sent from the slave site to the master site as a non-power signal to avoid passivity problems. Moreover, a nonlinear function is applied to each controller term as a smooth saturation function, providing a bounded control signal and preserving the system’s actuators. Lastly, the Lyapunov approach demonstrates the global stability of the controlled system, and numerical experiment results further confirm the validity of the presented strategy.

Keywords: teleoperation manipulators system, time-varying delay, actuator saturation, adaptive robust rbf neural network approximation, uncertainties

Procedia PDF Downloads 52

Authors: Ksenia Meshkova

Abstract:

With increasing amount of multimedia data, it is very important to automate and speed up the process of obtaining meta. This process means not just recognition of some object or its movement, but recognition of the entire scene versus separate frames and having timeline segmentation as a final result. Labeling datasets is time consuming, besides, attributing characteristics to particular scenes is clearly difficult due to their nature. In this article, we will consider autoencoders application to unsupervised scene recognition and clusterization based on interpretable features. Further, we will focus on particular types of auto encoders that relevant to our study. We will take a look at the specificity of deep learning related to information theory and rate-distortion theory and describe the solutions empowering poor interpretability of deep learning in media content processing. As a conclusion, we will present the results of the work of custom framework, based on autoencoders, capable of scene recognition as was deeply studied above, with highlights generation resulted out of this recognition. We will not describe in detail the mathematical description of neural networks work but will clarify the necessary concepts and pay attention to important nuances.

Keywords: neural networks, computer vision, representation learning, autoencoders

Procedia PDF Downloads 104

Authors: Ashish Payal, C. S. Rai, B. V. R. Reddy

Abstract:

With the increasing use and application of Wireless Sensor Networks (WSN), need has arisen to explore them in more effective and efficient manner. An important area which can bring efficiency to WSNs is the localization process, which refers to the estimation of the position of wireless sensor nodes in an ad hoc network setting, in reference to a coordinate system that may be internal or external to the network. In this paper, we have done comparison and analysed Sigmoidal Feedforward Artificial Neural Networks (SFFANNs) and Radial Basis Function (RBF) networks for developing localization framework in WSNs. The presented work utilizes the Received Signal Strength Indicator (RSSI), measured by static node on 100 x 100 m² grid from three anchor nodes. The comprehensive evaluation of these approaches is done using MATLAB software. The simulation results effectively demonstrate that FFANNs based sensor motes will show better localization accuracy as compared to RBF.

Keywords: localization, wireless sensor networks, artificial neural network, radial basis function, multi-layer perceptron, backpropagation, RSSI, GPS

Procedia PDF Downloads 313

Authors: Jianhong Xiang, Rui Sun, Linyu Wang

Abstract:

In recent years, due to the continuous improvement of deep learning theory, Convolutional Neural Network (CNN) has played a great superior performance in the research of Hyperspectral Image (HSI) classification. Since HSI has rich spatial-spectral information, only utilizing a single dimensional or single size convolutional kernel will limit the detailed feature information received by CNN, which limits the classification accuracy of HSI. In this paper, we design a multi-scale CNN with MLP based on residual and attention modules (INRAM-3DCNN) for the HSI classification task. We propose to use multiple 3D convolutional kernels to extract the packet feature information and fully learn the spatial-spectral features of HSI while designing residual 3D convolutional branches to avoid the decline of classification accuracy due to network degradation. Secondly, we also design the 2D Inception module with a joint channel attention mechanism to quickly extract key spatial feature information at different scales of HSI and reduce the complexity of the 3D model. Due to the high parallel processing capability and nonlinear global action of the Multilayer Perceptron (MLP), we use it in combination with the previous CNN structure for the final classification process. The experimental results on two HSI datasets show that the proposed INRAM-3DCNN method has superior classification performance and can perform the classification task excellently.

Keywords: INRAM-3DCNN, residual, channel attention, hyperspectral image classification

Procedia PDF Downloads 50

Authors: Dehini Rachid, Ferdi Brahim

Abstract:

The shunt active power filter (SAPF) is not destined only to improve the power factor, but also to compensate the unwanted harmonic currents produced by nonlinear loads. This paper presents a SAPF with identification and control method based on artificial neural network (ANN). To identify harmonics, many techniques are used, among them the conventional p-q theory and the relatively recent one the artificial neural network method. It is difficult to get satisfied identification and control characteristics by using a normal (ANN) due to the nonlinearity of the system (SAPF + fast nonlinear load variations). This work is an attempt to undertake a systematic study of the problem to equip the (SAPF) with the harmonics identification and DC link voltage control method based on (ANN). The latter has been applied to the (SAPF) with fast nonlinear load variations. The results of computer simulations and experiments are given, which can confirm the feasibility of the proposed active power filter.

Keywords: artificial neural networks (ANN), p-q theory, harmonics, total harmonic distortion

Procedia PDF Downloads 364

Authors: J. Balaraju, M. Govinda Raj, C. S. N. Murthy

Abstract:

Every industry is constantly looking for enhancement of its day to day production and productivity. This can be possible only by maintaining the men and machinery at its adequate level. Prediction of performance characteristics plays an important role in performance evaluation of the equipment. Analytical and statistical approaches will take a bit more time to solve complex problems such as performance estimations as compared with software-based approaches. Keeping this in view the present study deals with an Artificial Neural Network (ANN) modelling of a Load-Haul-Dump (LHD) machine to predict the performance characteristics such as reliability, availability and preventive maintenance (PM). A feed-forward-back-propagation ANN technique has been used to model the Levenberg-Marquardt (LM) training algorithm. The performance characteristics were computed using Isograph Reliability Workbench 13.0 software. These computed values were validated using predicted output responses of ANN models. Further, recommendations are given to the industry based on the performed analysis for improvement of equipment performance.

Keywords: load-haul-dump, LHD, artificial neural network, ANN, performance, reliability, availability, preventive maintenance

Procedia PDF Downloads 125

Authors: Tesfaw Belayneh Abebe

Abstract:

Integrating low-altitude earth orbit (LEO) satellites and unmanned aerial vehicles (UAVs) within a non-terrestrial network (NTN) with the assistance of reconfigurable intelligent surfaces (RIS), we investigate the problem of how to enhance throughput through integrated LEO satellites and UAVs with the assistance of RIS. We propose a method to jointly optimize the associations with the LEO satellite, the 3D trajectory of the UAV, and the phase shifts of the RIS to maximize communication throughput for RIS-assisted integrated LEO satellite and UAV-enabled wireless communications, which is challenging due to the time-varying changes in the position of the LEO satellite, the high mobility of UAVs, an enormous number of possible control actions, and also the large number of RIS elements. Utilizing a multi-agent double deep Q-network (MADDQN), our approach dynamically adjusts LEO satellite association, UAV positioning, and RIS phase shifts. Simulation results demonstrate that our method significantly outperforms baseline strategies in maximizing throughput. Lastly, thanks to the integrated network and the RIS, the proposed scheme achieves up to 65.66x higher peak throughput and 25.09x higher worst-case throughput.

Keywords: integrating low-altitude earth orbit (LEO) satellites, unmanned aerial vehicles (UAVs) within a non-terrestrial network (NTN), reconfigurable intelligent surfaces (RIS), multi-agent double deep Q-network (MADDQN)

Procedia PDF Downloads 19

Authors: Kemka Ihemelandu, Chukwuemeka Ihemelandu

Abstract:

Melanoma remains a public health crisis, with incidence rates increasing rapidly in the past decades. Improving diagnostic accuracy to decrease misdiagnosis using Artificial intelligence (AI) continues to be documented. Unfortunately, unintended racially biased outcomes, a product of lack of diversity in the dataset used, with a noted class imbalance favoring lighter vs. darker skin tone, have increasingly been recognized as a problem.Resulting in noted limitations of the accuracy of the Convolutional neural network (CNN)models. CNN models are prone to biased output due to biases in the dataset used to train them. Our aim in this study was the optimization of convolutional neural network algorithms to mitigate bias in the automated diagnosis of melanoma. We hypothesized that our proposed training algorithms based on a data augmentation method to optimize the diagnostic accuracy of a CNN classifier by generating new training samples from the original ones will reduce bias in the automated diagnosis of melanoma. We applied geometric transformation, including; rotations, translations, scale change, flipping, and shearing. Resulting in a CNN model that provided a modifiedinput data making for a model that could learn subtle racial features. Optimal selection of the momentum and batch hyperparameter increased our model accuracy. We show that our augmented model reduces bias while maintaining accuracy in the automated diagnosis of melanoma.

Keywords: bias, augmentation, melanoma, convolutional neural network

Procedia PDF Downloads 186

Authors: Liang Wang, Beste F. Yuksel, David Guy Brizan

Abstract:

A computational model of affect that can distinguish between spontaneous and posed smiles with no errors on a large, popular data set using deep learning techniques is presented in this paper. A Long Short-Term Memory (LSTM) classifier, a type of Recurrent Neural Network, is utilized and compared to human classification. Results showed that while human classification (mean of 0.7133) was above chance, the LSTM model was more accurate than human classification and other comparable state-of-the-art systems. Additionally, a high accuracy rate was maintained with small amounts of training videos (70 instances). The derivation of important features to further understand the success of our computational model were analyzed, and it was inferred that thousands of pairs of points within the eyes and mouth are important throughout all time segments in a smile. This suggests that distinguishing between a posed and spontaneous smile is a complex task, one which may account for the difficulty and lower accuracy of human classification compared to machine learning models.

Keywords: affective computing, affect detection, computer vision, deep learning, human-computer interaction, machine learning, posed smile detection, spontaneous smile detection

Procedia PDF Downloads 109

Authors: Martin K. Steiger, Lukas Heisler, Hans-Georg Brachtendorf

Abstract:

Deep neural networks and their variants form the backbone of many AI applications. Based on the so-called residual networks, a continuous formulation of such models as ordinary differential equations (ODEs) has proven advantageous since different techniques may be applied that significantly increase the learning speed and enable controlled trade-offs with the resulting error at the same time. For the evaluation of such models, high-performance numerical differential equation solvers are used, which also provide the gradients required for training. However, whether classical gradient-based methods are even applicable or which one yields the best results has not been discussed yet. This paper aims to redeem this situation by providing empirical results for different applications.

Keywords: deep neural networks, gradient-based learning, image processing, ordinary differential equation networks

Procedia PDF Downloads 141

Authors: Marcio Leal, Marta Villamil

Abstract:

Computacional recognition of sign languages aims to allow a greater social and digital inclusion of deaf people through interpretation of their language by computer. This article presents a model of recognition of two of global parameters from sign languages; hand configurations and hand movements. Hand motion is captured through an infrared technology and its joints are built into a virtual three-dimensional space. A Multilayer Perceptron Neural Network (MLP) was used to classify hand configurations and Dynamic Time Warping (DWT) recognizes hand motion. Beyond of the method of sign recognition, we provide a dataset of hand configurations and motion capture built with help of fluent professionals in sign languages. Despite this technology can be used to translate any sign from any signs dictionary, Brazilian Sign Language (Libras) was used as case study. Finally, the model presented in this paper achieved a recognition rate of 80.4%.

Keywords: artificial neural network, computer vision, dynamic time warping, infrared, sign language recognition

Procedia PDF Downloads 195

Authors: K. Bresilla, L. Manfrini, B. Morandi, A. Boini, G. Perulli, L. C. Grappadelli

Abstract:

Image/video processing for fruit in the tree using hard-coded feature extraction algorithms have shown high accuracy during recent years. While accurate, these approaches even with high-end hardware are computationally intensive and too slow for real-time systems. This paper details the use of deep convolution neural networks (CNNs), specifically an algorithm (YOLO - You Only Look Once) with 24+2 convolution layers. Using deep-learning techniques eliminated the need for hard-code specific features for specific fruit shapes, color and/or other attributes. This CNN is trained on more than 5000 images of apple and pear fruits on 960 cores GPU (Graphical Processing Unit). Testing set showed an accuracy of 90%. After this, trained data were transferred to an embedded device (Raspberry Pi gen.3) with camera for more portability. Based on correlation between number of visible fruits or detected fruits on one frame and the real number of fruits on one tree, a model was created to accommodate this error rate. Speed of processing and detection of the whole platform was higher than 40 frames per second. This speed is fast enough for any grasping/harvesting robotic arm or other real-time applications.

Keywords: artificial intelligence, computer vision, deep learning, fruit recognition, harvesting robot, precision agriculture

Procedia PDF Downloads 393

Authors: Alisher Ibragimov, Sofya Senotrusova, Aleksandra Beliaeva, Egor Ushakov, Yuri Markin

Abstract:

Mammography plays a critical role in screening for breast cancer in women, and artificial intelligence has enabled the automatic detection of diseases in medical images. Many of the current techniques used for mammogram analysis focus on a single view (mediolateral or craniocaudal view), while in clinical practice, radiologists consider multiple views of mammograms from both breasts to make a correct decision. Consequently, computer-aided diagnosis (CAD) systems could benefit from incorporating information gathered from multiple views. In this study, the introduce a method based on a Siamese neural network (SNN) model that simultaneously analyzes mammographic images from tri-view: bilateral and ipsilateral. In this way, when a decision is made on a single image of one breast, attention is also paid to two other images – a view of the same breast in a different projection and an image of the other breast as well. Consequently, the algorithm closely mimics the radiologist's practice of paying attention to the entire examination of a patient rather than to a single image. Additionally, to the best of our knowledge, this research represents the first experiments conducted using the recently released Vietnamese dataset of digital mammography (VinDr-Mammo). On an independent test set of images from this dataset, the best model achieved an AUC of 0.87 per image. Therefore, this suggests that there is a valuable automated second opinion in the interpretation of mammograms and breast cancer diagnosis, which in the future may help to alleviate the burden on radiologists and serve as an additional layer of verification.

Keywords: breast cancer, computer-aided diagnosis, deep learning, multi-view mammogram, siamese neural network

Procedia PDF Downloads 114

Authors: Reza Dinarvand, Mahdi Sadeghian, Somaye Sadeghian

Abstract:

Prediction of the ultimate bearing capacity of piles (Qu) is one of the basic issues in geotechnical engineering. So far, several methods have been used to estimate Qu, including the recently developed artificial intelligence methods. In recent years, optimization algorithms have been used to minimize artificial network errors, such as colony algorithms, genetic algorithms, imperialist competitive algorithms, and so on. In the present research, artificial neural networks based on colonial competition algorithm (ANN-ICA) were used, and their results were compared with other methods. The results of laboratory tests of short piles in clayey soils with parameters such as pile diameter, pile buried length, eccentricity of load and undrained shear resistance of soil were used for modeling and evaluation. The results showed that ICA-based artificial neural networks predicted lateral bearing capacity of short piles with a correlation coefficient of 0.9865 for training data and 0.975 for test data. Furthermore, the results of the model indicated the superiority of ICA-based artificial neural networks compared to back-propagation artificial neural networks as well as the Broms and Hansen methods.

Keywords: artificial neural network, clayey soil, imperialist competition algorithm, lateral bearing capacity, short pile

Procedia PDF Downloads 128

Authors: Gurmail Singh

Abstract:

Effective supply chain is the main priority of every organization which is the outcome of strategic corporate investments with deliberate management action. Value-driven supply chain is defined through development, procurement and by configuring the appropriate resources, metrics and processes. However, responsiveness of the supply chain can be improved by proper coordination. So the Bullwhip effect (BWE) and Net stock amplification (NSAmp) values were anticipated and used for the control of inventory in organizations by both discrete wavelet transform-Artificial neural network (DWT-ANN) and Adaptive Network-based fuzzy inference system (ANFIS). This work presents a comparative methodology of forecasting for the customers demand which is non linear in nature for a multilevel supply chain structure using hybrid techniques such as Artificial intelligence techniques including Artificial neural networks (ANN) and Adaptive Network-based fuzzy inference system (ANFIS) and Discrete wavelet theory (DWT). The productiveness of these forecasting models are shown by computing the data from real world problems for Bullwhip effect and Net stock amplification. The results showed that these parameters were comparatively less in case of discrete wavelet transform-Artificial neural network (DWT-ANN) model and using Adaptive network-based fuzzy inference system (ANFIS).

Keywords: bullwhip effect, hybrid techniques, net stock amplification, supply chain flexibility

Procedia PDF Downloads 104

Authors: Zachary Pan

Abstract:

Our general mission is to solve the problem of classifying images into different body part types and deciding if each of them is healthy or not. However, for now, we will determine healthiness for only one-sixth of the body parts, specifically the chest. We will detect pneumonia in X-ray scans of those chest images. With this type of AI, doctors can use it as a second opinion when they are taking CT or X-ray scans of their patients. Another ad-vantage of using this machine learning classifier is that it has no human weaknesses like fatigue. The overall ap-proach to this problem is to split the problem into two parts: first, classify the image, then determine if it is healthy. In order to classify the image into a specific body part class, the body parts dataset must be split into test and training sets. We can then use many models, like neural networks or logistic regression models, and fit them using the training set. Now, using the test set, we can obtain a realistic accuracy the models will have on images in the real world since these testing images have never been seen by the models before. In order to increase this testing accuracy, we can also apply many complex algorithms to the models, like multiplicative weight update. For the second part of the problem, to determine if the body part is healthy, we can have another dataset consisting of healthy and non-healthy images of the specific body part and once again split that into the test and training sets. We then use another neural network to train on those training set images and use the testing set to figure out its accuracy. We will do this process only for the chest images. A major conclusion reached is that convolutional neural networks are the most reliable and accurate at image classification. In classifying the images, the logistic regression model, the neural network, neural networks with multiplicative weight update, neural networks with the black box algorithm, and the convolutional neural network achieved 96.83 percent accuracy, 97.33 percent accuracy, 97.83 percent accuracy, 96.67 percent accuracy, and 98.83 percent accuracy, respectively. On the other hand, the overall accuracy of the model that de-termines if the images are healthy or not is around 78.37 percent accuracy.

Keywords: body part, healthcare, machine learning, neural networks

Procedia PDF Downloads 77