Search results for: deep feed forward neural network

Authors: Apoorva Vinod, Archana Mathur, Snehanshu Saha

Abstract:

Though there exists a plethora of Activation Functions (AFs) used in single and multiple hidden layer Neural Networks (NN), their behavior always raised curiosity, whether used in combination or singly. The popular AFs –Sigmoid, ReLU, and Tanh–have performed prominently well for shallow and deep architectures. Most of the time, AFs are used singly in multi-layered NN, and, to the best of our knowledge, their performance is never studied and analyzed deeply when used in combination. In this manuscript, we experiment with multi-layered NN architecture (both on shallow and deep architectures; Convolutional NN and VGG16) and investigate how well the network responds to using two different AFs (Sigmoid-Tanh, Tanh-ReLU, ReLU-Sigmoid) used alternately against a traditional, single (Sigmoid-Sigmoid, Tanh-Tanh, ReLUReLU) combination. Our results show that using two different AFs, the network achieves better accuracy, substantially lower loss, and faster convergence on 4 computer vision (CV) and 15 Non-CV (NCV) datasets. When using different AFs, not only was the accuracy greater by 6-7%, but we also accomplished convergence twice as fast. We present a case study to investigate the probability of networks suffering vanishing and exploding gradients when using two different AFs. Additionally, we theoretically showed that a composition of two or more AFs satisfies Universal Approximation Theorem (UAT).

Keywords: activation function, universal approximation function, neural networks, convergence

Procedia PDF Downloads 158

Authors: P. S. Jagadeesh Kumar, Yang Yung, Mingmin Pan, Xianpei Li, Wenli Hu

Abstract:

Glaucoma is impelled by optic nerve mutilation habitually represented as cupping and visual field injury frequently with an arcuate pattern of mid-peripheral loss, subordinate to retinal ganglion cell damage and death. Glaucoma is the second foremost cause of blindness and the chief cause of permanent blindness worldwide. Consequently, all-embracing study into the analysis and empathy of glaucoma is happening to escort deep learning based neural network intrusions to deliberate this substantial optic neuropathy. This paper advances an ophthalmic hashing based supervision of glaucoma and corneal disorders preeminent on deep graphical model. Ophthalmic hashing is a newly proposed method extending the efficacy of visual hash-coding to predict glaucoma corneal disorder matching, which is the faster than the existing methods. Deep graphical model is proficient of learning interior explications of corneal disorders in satisfactory time to solve hard combinatoric incongruities using deep Boltzmann machines.

Keywords: corneal disorders, deep Boltzmann machines, deep graphical model, glaucoma, neural networks, ophthalmic hashing

Procedia PDF Downloads 250

Authors: Lamyaa Gamal El-Deen Taha, Ashraf Sharawi

Abstract:

China launched satellite GF-2 in 2014. This study deals with comparing nearest neighbor object-based classification and neural network classification methods for classification of the fused GF-2 image. Firstly, rectification of GF-2 image was performed. Secondly, a comparison between nearest neighbor object-based classification and neural network classification for classification of fused GF-2 was performed. Thirdly, the overall accuracy of classification and kappa index were calculated. Results indicate that nearest neighbor object-based classification is better than neural network classification for urban mapping.

Keywords: GF-2 images, feature extraction-rectification, nearest neighbour object based classification, segmentation algorithms, neural network classification, multilayer perceptron

Procedia PDF Downloads 389

Authors: Jonathan Sohn

Abstract:

Deep neural networks (DNNs) have shown state-of-the-art performance for many applications, including computer vision, natural language processing, and speech recognition. Recently, adversarial attacks have been studied in the context of deep neural networks, which aim to alter the results of deep neural networks by modifying the inputs slightly. For example, an adversarial attack on a DNN used for object detection can cause the DNN to miss certain objects. As a result, the reliability of DNNs is undermined by their lack of robustness against adversarial attacks, raising concerns about their use in safety-critical applications such as autonomous driving. In this paper, we focus on studying the adversarial attacks and defenses on DNNs for image classification. There are two types of adversarial attacks studied which are fast gradient sign method (FGSM) attack and projected gradient descent (PGD) attack. A DNN forms decision boundaries that separate the input images into different categories. The adversarial attack slightly alters the image to move over the decision boundary, causing the DNN to misclassify the image. FGSM attack obtains the gradient with respect to the image and updates the image once based on the gradients to cross the decision boundary. PGD attack, instead of taking one big step, repeatedly modifies the input image with multiple small steps. There is also another type of attack called the target attack. This adversarial attack is designed to make the machine classify an image to a class chosen by the attacker. We can defend against adversarial attacks by incorporating adversarial examples in training. Specifically, instead of training the neural network with clean examples, we can explicitly let the neural network learn from the adversarial examples. In our experiments, the digit recognition accuracy on the MNIST dataset drops from 97.81% to 39.50% and 34.01% when the DNN is attacked by FGSM and PGD attacks, respectively. If we utilize FGSM training as a defense method, the classification accuracy greatly improves from 39.50% to 92.31% for FGSM attacks and from 34.01% to 75.63% for PGD attacks. To further improve the classification accuracy under adversarial attacks, we can also use a stronger PGD training method. PGD training improves the accuracy by 2.7% under FGSM attacks and 18.4% under PGD attacks over FGSM training. It is worth mentioning that both FGSM and PGD training do not affect the accuracy of clean images. In summary, we find that PGD attacks can greatly degrade the performance of DNNs, and PGD training is a very effective way to defend against such attacks. PGD attacks and defence are overall significantly more effective than FGSM methods.

Keywords: deep neural network, adversarial attack, adversarial defense, adversarial machine learning

Procedia PDF Downloads 194

Authors: Bryan Crompton, Daniel Giger, Tanay Mehta, Apurva Mody

Abstract:

The task of classifying radio frequency (RF) signals has seen recent success in employing deep neural network models. In this work, we present a combined signal processing and machine learning approach to signal classification for cochannel anomalous signals. The power spectral density and cyclostationary signal processing features of a captured signal are computed and fed into a neural net to produce a classification decision. Our combined signal preprocessing and machine learning approach allows for simpler neural networks with fast training times and small computational resource requirements for inference with longer preprocessing time.

Keywords: signal processing, machine learning, cyclostationary signal processing, signal classification

Procedia PDF Downloads 107

Authors: Halimatu S. Abdullahi, Ray E. Sheriff, Fatima Mahieddine

Abstract:

Agriculture is essential to the continuous existence of human life as they directly depend on it for the production of food. The exponential rise in population calls for a rapid increase in food with the application of technology to reduce the laborious work and maximize production. Technology can aid/improve agriculture in several ways through pre-planning and post-harvest by the use of computer vision technology through image processing to determine the soil nutrient composition, right amount, right time, right place application of farm input resources like fertilizers, herbicides, water, weed detection, early detection of pest and diseases etc. This is precision agriculture which is thought to be solution required to achieve our goals. There has been significant improvement in the area of image processing and data processing which has being a major challenge. A database of images is collected through remote sensing, analyzed and a model is developed to determine the right treatment plans for different crop types and different regions. Features of images from vegetations need to be extracted, classified, segmented and finally fed into the model. Different techniques have been applied to the processes from the use of neural network, support vector machine, fuzzy logic approach and recently, the most effective approach generating excellent results using the deep learning approach of convolution neural network for image classifications. Deep Convolution neural network is used to determine soil nutrients required in a plantation for maximum production. The experimental results on the developed model yielded results with an average accuracy of 99.58%.

Keywords: convolution, feature extraction, image analysis, validation, precision agriculture

Procedia PDF Downloads 315

Authors: Kemka C. Ihemelandu, Chukwuemeka U. Ihemelandu

Abstract:

The incidence of melanoma has been increasing rapidly over the past two decades, making melanoma a current public health crisis. Unfortunately, even as screening efforts continue to expand in an effort to ameliorate the death rate from melanoma, there is a need to improve diagnostic accuracy to decrease misdiagnosis. Artificial intelligence (AI) a new frontier in patient care has the ability to improve the accuracy of melanoma diagnosis. Convolutional neural network (CNN) a form of deep neural network, most commonly applied to analyze visual imagery, has been shown to outperform the human brain in pattern recognition. However, there are noted limitations with the accuracy of the CNN models. Our aim in this study was the optimization of convolutional neural network algorithms for the automated diagnosis of melanoma. We hypothesized that Optimal selection of the momentum and batch hyperparameter increases model accuracy. Our most successful model developed during this study, showed that optimal selection of momentum of 0.25, batch size of 2, led to a superior performance and a faster model training time, with an accuracy of ~ 83% after nine hours of training. We did notice a lack of diversity in the dataset used, with a noted class imbalance favoring lighter vs. darker skin tone. Training set image transformations did not result in a superior model performance in our study.

Keywords: melanoma, convolutional neural network, momentum, batch hyperparameter

Procedia PDF Downloads 101

Authors: Gautam Kumar Saharia, Sagardeep Talukdar, Riki Dutta, Sudipta Nandy

Abstract:

The physics-informed neural network (PINN) method opens up an approach for numerically solving nonlinear partial differential equations leveraging fast calculating speed and high precession of modern computing systems. We construct the PINN based on a strong universal approximation theorem and apply the initial-boundary value data and residual collocation points to weekly impose initial and boundary conditions to the neural network and choose the optimization algorithms adaptive moment estimation (ADAM) and Limited-memory Broyden-Fletcher-Golfard-Shanno (L-BFGS) algorithm to optimize learnable parameter of the neural network. Next, we improve the PINN with a weighted loss function to obtain both the bright and dark soliton solutions of the Fokas-Lenells equation (FLE). We find the proposed scheme of adjustable weight coefficients into PINN has a better convergence rate and generalizability than the basic PINN algorithm. We believe that the PINN approach to solve the partial differential equation appearing in nonlinear optics would be useful in studying various optical phenomena.

Keywords: deep learning, optical soliton, physics informed neural network, partial differential equation

Procedia PDF Downloads 70

Authors: Danladi Ali

Abstract:

In this paper, we measured GSM signal strength in the Dnepropetrovsk city in order to predict path loss in study area using nonlinear autoregressive neural network prediction and we also, used neural network clustering to determine average GSM signal strength receive at the study area. The nonlinear auto-regressive neural network predicted that the GSM signal is attenuated with the mean square error (MSE) of 2.6748dB, this attenuation value is used to modify the COST 231 Hata and the Okumura-Hata models. The neural network clustering revealed that -75dB to -95dB is received more frequently. This means that the signal strength received at the study is mostly weak signal

Keywords: one-dimensional multilevel wavelets, path loss, GSM signal strength, propagation, urban environment and model

Procedia PDF Downloads 382

Authors: Ilker Ali Ozkan

Abstract:

In this study, an artificial neural network model has been developed to estimate chronic kidney failure which is a common disease. The patients’ age, their blood and biochemical values, and 24 input data which consists of various chronic diseases are used for the estimation process. The input data have been subjected to preprocessing because they contain both missing values and nominal values. 147 patient data which was obtained from the preprocessing have been divided into as 70% training and 30% testing data. As a result of the study, artificial neural network model with 25 neurons in the hidden layer has been found as the model with the lowest error value. Chronic kidney failure disease has been able to be estimated accurately at the rate of 99.3% using this artificial neural network model. The developed artificial neural network has been found successful for the estimation of chronic kidney failure disease using clinical data.

Keywords: estimation, artificial neural network, chronic kidney failure disease, disease diagnosis

Procedia PDF Downloads 447

Authors: A. Suresh, Sreehari Rao Patri, K. S. R. Krishnaprasad

Abstract:

An ultra low power capacitor less low-dropout voltage regulator with improved transient response using gain enhanced feed forward path compensation is presented in this paper. It is based on a cascade of a voltage amplifier and a transconductor stage in the feed forward path with regular error amplifier to form a composite gain-enhanced feed forward stage. It broadens the gain bandwidth and thus improves the transient response without substantial increase in power consumption. The proposed LDO, designed for a maximum output current of 100 mA in UMC 180 nm, requires a quiescent current of 69 µA. An undershoot of 153.79mV for a load current changes from 0mA to 100mA and an overshoot of 196.24mV for current change of 100mA to 0mA. The settling time is approximately 1.1 µs for the output voltage undershoot case. The load regulation is of 2.77 µV/mA at load current of 100mA. Reference voltage is generated by using an accurate band gap reference circuit of 0.8V.The costly features of SOC such as total chip area and power consumption is drastically reduced by the use of only a total compensation capacitance of 6pF while consuming power consumption of 0.096 mW.

Keywords: capacitor-less LDO, frequency compensation, transient response, latch, self-biased differential amplifier

Procedia PDF Downloads 450

Authors: Aydin Teymourifar, Gurkan Ozturk

Abstract:

In this paper, a new hybrid Hopfield neural network is proposed for the dynamic, flexible job shop scheduling problem. A new heuristic based and easy to implement energy function is designed for the Hopfield neural network, which penalizes the constraints violation and decreases makespan. Moreover, for enhancing the performance, several heuristics are integrated to it that achieve active, and non-delay schedules also, prevent early convergence of the neural network. The suggested algorithm that is designed as a generalization of the previous studies for the flexible and dynamic scheduling problems can be used for solving real scheduling problems. Comparison of the presented hybrid method results with the previous studies results proves its efficiency.

Keywords: dynamic flexible job shop scheduling, neural network, heuristics, constrained optimization

Procedia PDF Downloads 418

Authors: Farshid Hajati, Federico Girosi, Shima Ghassempour

Abstract:

We present a Deep Belief Network (DBN) method for clustering health trajectories. Deep Belief Network (DBN) is a deep architecture that consists of a stack of Restricted Boltzmann Machines (RBM). In a deep architecture, each layer learns more complex features than the past layers. The proposed method depends on DBN in clustering without using back propagation learning algorithm. The proposed DBN has a better a performance compared to the deep neural network due the initialization of the connecting weights. We use Contrastive Divergence (CD) method for training the RBMs which increases the performance of the network. The performance of the proposed method is evaluated extensively on the Health and Retirement Study (HRS) database. The University of Michigan Health and Retirement Study (HRS) is a nationally representative longitudinal study that has surveyed more than 27,000 elderly and near-elderly Americans since its inception in 1992. Participants are interviewed every two years and they collect data on physical and mental health, insurance coverage, financial status, family support systems, labor market status, and retirement planning. The dataset is publicly available and we use the RAND HRS version L, which is easy to use and cleaned up version of the data. The size of sample data set is 268 and the length of the trajectories is equal to 10. The trajectories do not stop when the patient dies and represent 10 different interviews of live patients. Compared to the state-of-the-art benchmarks, the experimental results show the effectiveness and superiority of the proposed method in clustering health trajectories.

Keywords: health trajectory, clustering, deep learning, DBN

Procedia PDF Downloads 369

Authors: Li Hui, Riyadh Hindi

Abstract:

Concrete structures frequently suffer from crack formation, a critical issue that can significantly reduce their lifespan by allowing damaging agents to enter. Traditional methods of crack detection depend on manual visual inspections, which heavily relies on the experience and expertise of inspectors using tools. In this study, a more efficient, computer vision-based approach is introduced by using the lightweight MobileNetV2 neural network. A dataset of 40,000 images was used to develop a specialized crack evaluation algorithm. The analysis indicates that MobileNetV2 matches the accuracy of traditional CNN methods but is more efficient due to its smaller size, making it well-suited for mobile device applications. The effectiveness and reliability of this new method were validated through experimental testing, highlighting its potential as an automated solution for crack detection in concrete structures.

Keywords: Concrete crack, computer vision, deep learning, MobileNetV2 neural network

Procedia PDF Downloads 66

Authors: Selim M. Khan

Abstract:

Indoor air quality is strongly influenced by the presence of radioactive radon (222Rn) gas. Indeed, exposure to high 222Rn concentrations is unequivocally linked to DNA damage and lung cancer and is a worsening issue in North American and European built environments, having increased over time within newer housing stocks as a function of as yet unclear variables. Indoor air radon concentration can be influenced by a wide range of environmental, structural, and behavioral factors. As some of these factors are quantitative while others are qualitative, no single statistical model can determine indoor radon level precisely while simultaneously considering all these variables across a complex and highly diverse dataset. The ability of AI- machine (deep) learning to simultaneously analyze multiple quantitative and qualitative features makes it suitable to predict radon with a high degree of precision. Using Canadian and Swedish long-term indoor air radon exposure data, we are using artificial deep neural network models with random weights and polynomial statistical models in MATLAB to assess and predict radon health risk to human as a function of geospatial, human behavioral, and built environmental metrics. Our initial artificial neural network with random weights model run by sigmoid activation tested different combinations of variables and showed the highest prediction accuracy (>96%) within the reasonable iterations. Here, we present details of these emerging methods and discuss strengths and weaknesses compared to the traditional artificial neural network and statistical methods commonly used to predict indoor air quality in different countries. We propose an artificial deep neural network with random weights as a highly effective method for assessing and predicting indoor radon.

Keywords: radon, radiation protection, lung cancer, aI-machine deep learnng, risk assessment, risk prediction, Europe, North America

Procedia PDF Downloads 96

Authors: Hassan Masoumi, Ahad Salimi, Nazanin Barhemmat, Babak Gholami

Abstract:

Artificial neural networks have gained a lot of interest as empirical models for their powerful representational capacity, multi input and output mapping characteristics. In fact, most feed-forward networks with nonlinear nodal functions have been proved to be universal approximates. In this paper, we propose a new supervised method for color image classification based on self organizing feature maps (SOFM). This algorithm is based on competitive learning. The method partitions the input space using self-organizing feature maps to introduce the concept of local neighborhoods. Our image classification system entered into RGB image. Experiments with simulated data showed that separability of classes increased when increasing training time. In additional, the result shows proposed algorithms are effective for color image classification.

Keywords: classification, SOFM algorithm, neural network, neighborhood, RGB image

Procedia PDF Downloads 478

Authors: Ashutosh Mishra, Nikhil Goyal

Abstract:

A 3D facial emotion recognition model based on deep learning is proposed in this paper. Two convolution layers and a pooling layer are employed in the deep learning architecture. After the convolution process, the pooling is finished. The probabilities for various classes of human faces are calculated using the sigmoid activation function. To verify the efficiency of deep learning-based systems, a set of faces. The Kaggle dataset is used to verify the accuracy of a deep learning-based face recognition model. The model's accuracy is about 65 percent, which is lower than that of other facial expression recognition techniques. Despite significant gains in representation precision due to the nonlinearity of profound image representations.

Keywords: facial recognition, computational intelligence, convolutional neural network, depth map

Procedia PDF Downloads 231

Authors: Li Zhang, Yuehong Su

Abstract:

Lightpipe as a mature solar light tube technique has been employed worldwide. Accurately assessing the performance of lightpipe and evaluate daylighting available has been a challenging topic. Previous research had used regression model and computational simulation methods to estimate the performance of lightpipe. However, due to the nonlinear nature of solar light transferring in lightpipe, the methods mentioned above express inaccurate and time-costing issues. In the present study, a neural network model as an alternative method is investigated to predict the transmittance of lightpipe. Four types of commercial lightpipe with bended angle 0°, 30°, 45° and 60° are discussed under clear, intermediate and overcast sky conditions respectively. The neural network is generated in MATLAB by using the outcomes of an optical software Photopia simulations as targets for networks training and testing. The coefficient of determination (R²) for each model is higher than 0.98, and the mean square error (MSE) is less than 0.0019, which indicate the neural network strong predictive ability and the use of the neural network method could be an efficient technique for determining the performance of lightpipe.

Keywords: neural network, bended lightpipe, transmittance, Photopia

Procedia PDF Downloads 152

Authors: Jiangxia Han, Liang Xue, Keda Chen

Abstract:

Despite the significant progress over the last decades in reservoir simulation using numerical discretization, meshing is complex. Moreover, the high degree of freedom of the space-time flow field makes the solution process very time-consuming. Therefore, we present Physics-Informed Convolutional Neural Networks(PICNN) as a hybrid scientific theory and data method for reservoir modeling. Besides labeled data, the model is driven by the scientific theories of the underlying problem, such as governing equations, boundary conditions, and initial conditions. PICNN integrates governing equations and boundary conditions into the network architecture in the form of a customized convolution kernel. The loss function is composed of data matching, initial conditions, and other measurable prior knowledge. By customizing the convolution kernel and minimizing the loss function, the neural network parameters not only fit the data but also honor the governing equation. The PICNN provides a methodology to model and history-match flow and transport problems in porous media. Numerical results demonstrate that the proposed PICNN can provide an accurate physical solution from a limited dataset. We show how this method can be applied in the context of a forward simulation for continuous problems. Furthermore, several complex scenarios are tested, including the existence of data noise, different work schedules, and different good patterns.

Keywords: convolutional neural networks, deep learning, flow and transport in porous media, physics-informed neural networks, reservoir simulation

Procedia PDF Downloads 143

Authors: William Whiteley, Jens Gregor

Abstract:

In recent developments of direct neural network based positron emission tomography (PET) reconstruction, two prominent architectures have emerged for converting measurement data into images: 1) networks that contain fully-connected layers; and 2) networks that primarily use a convolutional encoder-decoder architecture. In this paper, we present a multi-modal direct PET reconstruction method called MDPET, which is a hybrid approach that combines the advantages of both types of networks. MDPET processes raw data in the form of sinograms and histo-images in concert with attenuation maps to produce high quality multi-slice PET images (e.g., 8x440x440). MDPET is trained on a large whole-body patient data set and evaluated both quantitatively and qualitatively against target images reconstructed with the standard PET reconstruction benchmark of iterative ordered subsets expectation maximization. The results show that MDPET outperforms the best previously published direct neural network methods in measures of bias, signal-to-noise ratio, mean absolute error, and structural similarity.

Keywords: deep learning, image reconstruction, machine learning, neural network, positron emission tomography

Procedia PDF Downloads 110

Authors: Mo-Se Lee, Cheol-Hwi Ahn, Kee-Young Kwahk, Hyunchul Ahn

Abstract:

Over the past decade, deep learning has been in spotlight among various machine learning algorithms. In particular, CNN (Convolutional Neural Network), which is known as effective solution for recognizing and classifying images, has been popularly applied to classification and prediction problems in various fields. In this study, we try to apply CNN to stock market prediction, one of the most challenging tasks in the machine learning research. In specific, we propose to apply CNN as the binary classifier that predicts stock market direction (up or down) by using a graph as its input. That is, our proposal is to build a machine learning algorithm that mimics a person who looks at the graph and predicts whether the trend will go up or down. Our proposed model consists of four steps. In the first step, it divides the dataset into 5 days, 10 days, 15 days, and 20 days. And then, it creates graphs for each interval in step 2. In the next step, CNN classifiers are trained using the graphs generated in the previous step. In step 4, it optimizes the hyper parameters of the trained model by using the validation dataset. To validate our model, we will apply it to the prediction of KOSPI200 for 1,986 days in eight years (from 2009 to 2016). The experimental dataset will include 14 technical indicators such as CCI, Momentum, ROC and daily closing price of KOSPI200 of Korean stock market.

Keywords: convolutional neural network, deep learning, Korean stock market, stock market prediction

Procedia PDF Downloads 425

Authors: Sello Mokwena, Monyepao Thabang

Abstract:

In this study, we propose a comprehensive approach to address the challenges in person re-identification models. By combining a centroid tracking algorithm with a Siamese convolutional neural network model, our method excels in detecting, tracking, and capturing robust person features across non-overlapping camera views. The algorithm efficiently identifies individuals in the camera network, while the neural network extracts fine-grained global features for precise cross-image comparisons. The approach's effectiveness is further accentuated by leveraging the camera network topology for guidance. Our empirical analysis on benchmark datasets highlights its competitive performance, particularly evident when background subtraction techniques are selectively applied, underscoring its potential in advancing person re-identification techniques.

Keywords: camera network, convolutional neural network topology, person tracking, person re-identification, siamese

Procedia PDF Downloads 72

Authors: Megha Gupta, Nupur Prakash

Abstract:

Identification of plant diseases has been performed using machine learning and deep learning models on the datasets containing images of healthy and diseased plant leaves. The current study carries out an evaluation of some of the deep learning models based on convolutional neural network (CNN) architectures for identification of plant diseases. For this purpose, the publicly available New Plant Diseases Dataset, an augmented version of PlantVillage dataset, available on Kaggle platform, containing 87,900 images has been used. The dataset contained images of 26 diseases of 14 different plants and images of 12 healthy plants. The CNN models selected for the study presented in this paper are AlexNet, ZFNet, VGGNet (four models), GoogLeNet, and ResNet (three models). The selected models are trained using PyTorch, an open-source machine learning library, on Google Colaboratory. A comparative study has been carried out to analyze the high degree of accuracy achieved using these models. The highest test accuracy and F1-score of 99.59% and 0.996, respectively, were achieved by using GoogLeNet with Mini-batch momentum based gradient descent learning algorithm.

Keywords: comparative analysis, convolutional neural networks, deep learning, plant disease identification

Procedia PDF Downloads 198

Authors: Lizhi Ma, Dan Liu

Abstract:

Pipeline parallelism has been widely used to accelerate distributed deep learning to alleviate GPU memory bottlenecks and to ensure that models can be trained and deployed smoothly under limited graphics memory conditions. However, in highly heterogeneous distributed clusters, traditional model partitioning methods are not able to achieve load balancing. The overlap of communication and computation is also a big challenge. In this paper, HePipe is proposed, an efficient pipeline parallel training method for highly heterogeneous clusters. According to the characteristics of the neural network model pipeline training task, oriented to the 2-level heterogeneous cluster computing topology, a training method based on the 2-level stage division of neural network modeling and partitioning is designed to improve the parallelism. Additionally, a multi-forward 1F1B scheduling strategy is designed to accelerate the training time of each stage by executing the computation units in advance to maximize the overlap between the forward propagation communication and backward propagation computation. Finally, a dynamic recomputation strategy based on task memory requirement prediction is proposed to improve the fitness ratio of task and memory, which improves the throughput of the cluster and solves the memory shortfall problem caused by memory differences in heterogeneous clusters. The empirical results show that HePipe improves the training speed by 1.6×−2.2× over the existing asynchronous pipeline baselines.

Keywords: pipeline parallelism, heterogeneous cluster, model training, 2-level stage partitioning

Procedia PDF Downloads 18

Authors: Arya Prakash Padhi, Anupam Chakrabarti, Rajib Chowdhury

Abstract:

Topology or layout optimization often gives better performing economic structures and is very helpful in the conceptual design phase. But traditionally it is being done in finite element-based optimization schemes which, although gives a good result, is very time-consuming especially in 3D structures. Among other alternatives machine learning, especially deep learning-based methods, have a very good potential in resolving this computational issue. Here convolutional neural network (3D-CNN) based variational auto encoder (VAE) is trained using a dataset generated from commercially available topology optimization code ABAQUS Tosca using solid isotropic material with penalization (SIMP) method for compliance minimization. The encoded data in latent space is then fed to a 3D generative adversarial network (3D-GAN) to generate the outcome in 64x64x64 size. Here the network consists of 3D volumetric CNN with rectified linear unit (ReLU) activation in between and sigmoid activation in the end. The proposed network is seen to provide almost optimal results with significantly reduced computational time, as there is no iteration involved.

Keywords: 3D generative adversarial network, deep learning, structural topology optimization, variational auto encoder

Procedia PDF Downloads 174

Authors: Vijaya Yuvaram Singh V M, Kameshwar Rao J V

Abstract:

The challenge with development of neural network based methods for medical is the availability of data. Anatomical landmark detection in the medical domain is a process to find points on the x-ray scan report of the patient. Most of the time this task is done manually by trained professionals as it requires precision and domain knowledge. Traditionally object detection based methods are used for landmark detection. Here, we utilize reinforcement learning and query based method to train a single agent capable of detecting multiple landmarks. A deep Q network agent is trained to detect single and multiple landmarks present on hip and shoulder from x-ray scan of a patient. Here a single agent is trained to find multiple landmark making it superior to having individual agents per landmark. For the initial study, five images of different patients are used as the environment and tested the agents performance on two unseen images.

Keywords: reinforcement learning, medical landmark detection, multi target detection, deep neural network

Procedia PDF Downloads 142

Authors: Joseph Uzuazomaro Oju

Abstract:

The ease at which manipulated media can be created, and the increasing difficulty in identifying fake media makes it a great threat. Most of the applications used for the creation of these high-quality fake videos and images are built with deep learning. Hence, the use of deep learning in creating a detection mechanism cannot be overemphasized. Any successful fake media that is being detected before it reached the populace will save people from the self-doubt of either a content is genuine or fake and will ensure the credibility of videos and images. The methodology introduced in this paper approaches the manipulated media detection challenge using a combo of VGG-19 and a deep capsule network. In the case of videos, they are converted into frames, which, in turn, are resized and cropped to the face region. These preprocessed images/videos are fed to the VGG-19 network to extract the latent features. The extracted latent features are inputted into a deep capsule network enhanced with a 3D -convolution dynamic routing agreement. The 3D –convolution dynamic routing agreement algorithm helps to reduce the linkages between capsules networks. Thereby limiting the poor learning shortcoming of multiple capsule network layers. The resultant output from the deep capsule network will indicate a media to be either genuine or fake.

Keywords: deep capsule network, dynamic routing, fake media detection, manipulated media

Procedia PDF Downloads 132

Authors: Goodness Onwuka, Khaled Abou-El-Hossein

Abstract:

Borosilicate-crown (BK7) glass has found broad application in the optic and automotive industries and the growing demands for nanometric surface finishes is becoming a necessity in such applications. Thus, it has become paramount to optimize the parameters influencing the surface roughness of this precision lens. The research was carried out on a 4-axes Nanoform 250 precision lathe machine with an ultra-high precision grinding spindle. The experiment varied the machining parameters of feed rate, wheel speed and depth of cut at three levels for different combinations using Box Behnken design of experiment and the resulting surface roughness values were measured using a Taylor Hobson Dimension XL optical profiler. Acoustic emission monitoring technique was applied at a high sampling rate to monitor the machining process while further signal processing and feature extraction methods were implemented to generate the input to a neural network algorithm. This paper highlights the training and development of a back propagation neural network prediction algorithm through careful selection of parameters and the result show a better classification accuracy when compared to a previously developed response surface model with very similar machining parameters. Hence artificial neural network algorithms provide better surface roughness prediction accuracy in the ultra-high precision grinding of BK7 glass.

Keywords: acoustic emission technique, artificial neural network, surface roughness, ultra-high precision grinding

Procedia PDF Downloads 305

Authors: Youji Miyake

Abstract:

In aircraft assembly, a large number of preparatory holes are required for screw and rivet joints. Currently, many holes are drilled manually because it is difficult to machine the holes using conventional computerized numerical control(CNC) machines. The application of industrial robots to drill the hole has been considered as an alternative to the CNC machines. However, the rigidity of robot arms is so low that vibration is likely to occur during drilling. In this study, it is proposed constant-load feed machining as a method to perform high-precision drilling while minimizing the thrust force, which is considered to be the cause of vibration. In this method, the drill feed is realized by a constant load applied onto the tool so that the thrust force is theoretically kept below the applied load. The performance of the proposed method was experimentally examined through the deep hole drilling of plastic and simultaneous drilling of metal/plastic stack plates. It was confirmed that the deep hole drilling and simultaneous drilling could be performed without generating vibration by controlling the tool feed rate in the appropriate range.

Keywords: constant load feed machining, robotic drilling, deep hole, simultaneous drilling

Procedia PDF Downloads 194

Authors: Liu Zhiyuan, Sun Zongdi

Abstract:

In this paper, the BP neural network model is established to predict the carbon trading price and carbon trading volume in Shanghai City. First of all, we find the data of carbon trading price and carbon trading volume in Shanghai City from September 30, 2015 to December 23, 2016. The carbon trading price and trading volume data were processed to get the average value of each 5, 10, 20, 30, and 60 carbon trading price and trading volume. Then, these data are used as input of BP neural network model. Finally, after the training of BP neural network, the prediction values of Shanghai carbon trading price and trading volume are obtained, and the model is tested.

Keywords: Carbon trading price, carbon trading volume, BP neural network model, Shanghai City

Procedia PDF Downloads 352