Search results for: deep convolution networks

Authors: Shiuh-Jer Huang, Chen-Zon Yan, C. K. Huang, Chun-Chien Ting

Abstract:

Since the vision system application in industrial environment for autonomous purposes is required intensely, the image recognition technique becomes an important research topic. Here, deep learning algorithm is employed in image system to recognize the industrial object and integrate with a 7A6 Series Manipulator for object automatic gripping task. PC and Graphic Processing Unit (GPU) are chosen to construct the 3D Vision Recognition System. Depth Camera (Intel RealSense SR300) is employed to extract the image for object recognition and coordinate derivation. The YOLOv2 scheme is adopted in Convolution neural network (CNN) structure for object classification and center point prediction. Additionally, image processing strategy is used to find the object contour for calculating the object orientation angle. Then, the specified object location and orientation information are sent to robotic controller. Finally, a six-axis manipulator can grasp the specific object in a random environment based on the user command and the extracted image information. The experimental results show that YOLOv2 has been successfully employed to detect the object location and category with confidence near 0.9 and 3D position error less than 0.4 mm. It is useful for future intelligent robotic application in industrial 4.0 environment.

Keywords: deep learning, image processing, convolution neural network, YOLOv2, 7A6 series manipulator

Procedia PDF Downloads 249

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 167

Authors: Shuen-Tai Wang, Fang-An Kuo, Chau-Yi Chou, Yu-Bin Fang

Abstract:

2016 has become the year of the Artificial Intelligence explosion. AI technologies are getting more and more matured that most world well-known tech giants are making large investment to increase the capabilities in AI. Machine learning is the science of getting computers to act without being explicitly programmed, and deep learning is a subset of machine learning that uses deep neural network to train a machine to learn  features directly from data. Deep learning realizes many machine learning applications which expand the field of AI. At the present time, deep learning frameworks have been widely deployed on servers for deep learning applications in both academia and industry. In training deep neural networks, there are many standard processes or algorithms, but the performance of different frameworks might be different. In this paper we evaluate the running performance of two state-of-the-art distributed deep learning frameworks that are running training calculation in parallel over multi GPU and multi nodes in our cloud environment. We evaluate the training performance of the frameworks with ResNet-50 convolutional neural network, and we analyze what factors that result in the performance among both distributed frameworks as well. Through the experimental analysis, we identify the overheads which could be further optimized. The main contribution is that the evaluation results provide further optimization directions in both performance tuning and algorithmic design.

Keywords: artificial intelligence, machine learning, deep learning, convolutional neural networks

Procedia PDF Downloads 210

Authors: Romisaa Ali

Abstract:

The Policy Gradient approach is one of the deep reinforcement learning families that combines deep neural networks (DNN) with reinforcement learning RL to discover the optimum of the control problem through experience gained from the interaction between the robot and its surroundings. In contrast to earlier policy gradient algorithms, which were unable to handle these two types of error because of over-or under-estimation introduced by the deep neural network model, this article will discuss the state-of-the-art SOTA policy gradient technique, trust region policy optimization (TRPO), by applying this method in various environments compared to another policy gradient method, the Proximal Policy Optimization (PPO), to explain their robust optimization, using this SOTA to gather experience data during various training phases after observing the impact of hyper-parameters on neural network performance.

Keywords: deep neural networks, deep reinforcement learning, proximal policy optimization, state-of-the-art, trust region policy optimization

Procedia PDF Downloads 168

Authors: Qingqing Gai, Junxing Shen, Yu Luo

Abstract:

Due to the hardware architecture characteristics, some operators tend to acquire better performance if the input/output tensor dimensions are aligned to a certain minimum granularity, such as convolution and deconvolution commonly used in deep learning. Furthermore, if the requirements are not met, the general strategy is to pad with 0 to satisfy the requirements, potentially leading to the under-utilization of the hardware resources. Therefore, for the convolution and deconvolution whose input and output channels do not meet the minimum granularity alignment, we propose to transfer the W-dimensional data to the C-dimension for computation (W2C) to enable the C-dimension to meet the hardware requirements. This scheme also reduces the number of computations in the W-dimension. Although this scheme substantially increases computation, the operator’s speed can improve significantly. It achieves remarkable speedups on multiple hardware accelerators, including Nvidia Tensor cores, Qualcomm digital signal processors (DSPs), and Huawei neural processing units (NPUs). All you need to do is modify the network structure and rearrange the operator weights offline without retraining. At the same time, for some operators, such as the Reducemax, we observe that transferring the Cdimensional data to the W-dimension(C2W) and replacing the Reducemax with the Maxpool can accomplish acceleration under certain circumstances.

Keywords: convolution, deconvolution, W2C, C2W, alignment, hardware accelerator

Procedia PDF Downloads 104

Authors: Yasser F. Hassan

Abstract:

Deep learning structure is a branch of machine learning science and greet achievement in research and applications. Cellular neural networks are regarded as array of nonlinear analog processors called cells connected in a way allowing parallel computations. The paper discusses how to use deep learning structure for representing neural cellular automata model. The proposed learning technique in cellular automata model will be examined from structure of deep learning. A deep automata neural cellular system modifies each neuron based on the behavior of the individual and its decision as a result of multi-level deep structure learning. The paper will present the architecture of the model and the results of simulation of approach are given. Results from the implementation enrich deep neural cellular automata system and shed a light on concept formulation of the model and the learning in it.

Keywords: cellular automata, neural cellular automata, deep learning, classification

Procedia PDF Downloads 196

Authors: Aishrith Rao

Abstract:

COVID 19 is a pandemic which has resulted in thousands of deaths around the world and a huge impact on the global economy. Testing is a huge issue as the test kits have limited availability and are expensive to manufacture. Using deep learning methods on radiology images in the detection of the coronavirus as these images contain information about the spread of the virus in the lungs is extremely economical and time-saving as it can be used in areas with a lack of testing facilities. This paper focuses on binary classification and multi-class classification of COVID 19 and other diseases such as pneumonia, tuberculosis, etc. Different deep learning methods such as VGG-19, COVID-Net, ResNET+ SVM, Deep CNN, DarkCovidnet, etc., have been used, and their accuracy has been compared using the Chest X-Ray dataset.

Keywords: deep learning, computer vision, radiology, COVID-19, ResNet, VGG-19, deep neural networks

Procedia PDF Downloads 159

Authors: Divvela Vishnu Sai Kumar, Deepak Arora, Sheenu Rizvi

Abstract:

Forecasting a cyclone through satellite images consists of the estimation of the intensity of the cyclone and predicting it before a cyclone comes. This research work can help people to take safety measures before the cyclone comes. The prediction of the intensity of a cyclone is very important to save lives and minimize the damage caused by cyclones. These cyclones are very costliest natural disasters that cause a lot of damage globally due to a lot of hazards. Authors have proposed five different CNN (Convolutional Neural Network) models that estimate the intensity of cyclones through INSAT-3D IR images. There are a lot of techniques that are used to estimate the intensity; the best model proposed by authors estimates intensity with a root mean squared error (RMSE) of 10.02 kts.

Keywords: estimating cyclone intensity, deep learning, convolution neural network, prediction models

Procedia PDF Downloads 125

Authors: Ja-Keoung Koo, Kensuke Nakamura, Hyohun Kim, Dongwha Shin, Yeonseok Kim, Ji-Su Ahn, Byung-Woo Hong

Abstract:

The machine learning techniques based on a convolutional neural network (CNN) have been actively developed and successfully applied to a variety of image analysis tasks including reconstruction, noise reduction, resolution enhancement, segmentation, motion estimation, object recognition. The classical visual information processing that ranges from low level tasks to high level ones has been widely developed in the deep learning framework. It is generally considered as a challenging problem to derive visual interpretation from high dimensional imagery data. A CNN is a class of feed-forward artificial neural network that usually consists of deep layers the connections of which are established by a series of non-linear operations. The CNN architecture is known to be shift invariant due to its shared weights and translation invariance characteristics. However, it is often computationally intractable to optimize the network in particular with a large number of convolution layers due to a large number of unknowns to be optimized with respect to the training set that is generally required to be large enough to effectively generalize the model under consideration. It is also necessary to limit the size of convolution kernels due to the computational expense despite of the recent development of effective parallel processing machinery, which leads to the use of the constantly small size of the convolution kernels throughout the deep CNN architecture. However, it is often desired to consider different scales in the analysis of visual features at different layers in the network. Thus, we propose a CNN model where different sizes of the convolution kernels are applied at each layer based on the random projection. We apply random filters with varying sizes and associate the filter responses with scalar weights that correspond to the standard deviation of the random filters. We are allowed to use large number of random filters with the cost of one scalar unknown for each filter. The computational cost in the back-propagation procedure does not increase with the larger size of the filters even though the additional computational cost is required in the computation of convolution in the feed-forward procedure. The use of random kernels with varying sizes allows to effectively analyze image features at multiple scales leading to a better generalization. The robustness and effectiveness of the proposed CNN based on random kernels are demonstrated by numerical experiments where the quantitative comparison of the well-known CNN architectures and our models that simply replace the convolution kernels with the random filters is performed. The experimental results indicate that our model achieves better performance with less number of unknown weights. The proposed algorithm has a high potential in the application of a variety of visual tasks based on the CNN framework. Acknowledgement—This work was supported by the MISP (Ministry of Science and ICT), Korea, under the National Program for Excellence in SW (20170001000011001) supervised by IITP, and NRF-2014R1A2A1A11051941, NRF2017R1A2B4006023.

Keywords: deep learning, convolutional neural network, random kernel, random projection, dimensionality reduction, object recognition

Procedia PDF Downloads 286

Authors: Shibo Wei, Ting Jiang

Abstract:

Convolutional-recurrent neural networks (CRN) have achieved much success recently in the speech enhancement field. The common processing method is to use the convolution layer to compress the feature space by multiple upsampling and then model the compressed features with the LSTM layer. At last, the enhanced speech is obtained by deconvolution operation to integrate the global information of the speech sequence. However, the feature space compression process may cause the loss of information, so we propose to model the upsampling result of each step with the residual LSTM layer, then join it with the output of the deconvolution layer and input them to the next deconvolution layer, by this way, we want to integrate the global information of speech sequence better. The experimental results show the network model (RES-CRN) we introduce can achieve better performance than LSTM without residual and overlaying LSTM simply in the original CRN in terms of scale-invariant signal-to-distortion ratio (SI-SNR), speech quality (PESQ), and intelligibility (STOI).

Keywords: convolutional-recurrent neural networks, speech enhancement, residual LSTM, SI-SNR

Procedia PDF Downloads 198

Authors: Farshad Zeinali, Sajedeh Norouzi, Nader Mokari, Eduard Jorswieck

Abstract:

The capacity of fifth-generation (5G) vehicle-to-everything (V2X) networks poses significant challenges. To ad- dress this challenge, this paper utilizes New Radio (NR) and New Radio Unlicensed (NR-U) networks to develop a heterogeneous vehicular network (HetNet). We propose a new framework, named joint BS assignment and resource allocation (JBSRA) for mobile V2X users and also consider coexistence schemes based on flexible duty cycle (DC) mechanism for unlicensed bands. Our objective is to maximize the average throughput of vehicles while guaranteeing the WiFi users' throughput. In simulations based on deep reinforcement learning (DRL) algorithms such as deep deterministic policy gradient (DDPG) and deep Q network (DQN), our proposed framework outperforms existing solutions that rely on fixed DC or schemes without consideration of unlicensed bands.

Keywords: vehicle-to-everything (V2X), resource allocation, BS assignment, new radio (NR), new radio unlicensed (NR-U), coexistence NR-U and WiFi, deep deterministic policy gradient (DDPG), deep Q-network (DQN), joint BS assignment and resource allocation (JBSRA), duty cycle mechanism

Procedia PDF Downloads 102

Authors: Muhammad Zubair Khan, Yugyung Lee

Abstract:

Deep learning has recently achieved enormous response in semantic image segmentation. The previously developed U-Net inspired architectures operate with continuous stride and pooling operations, leading to spatial data loss. Also, the methods lack establishing long-term pixels connection to preserve context knowledge and reduce spatial loss in prediction. This article developed encoder-decoder architecture with bi-directional LSTM embedded in long skip-connections and densely connected convolution blocks. The network non-linearly combines the feature maps across encoder-decoder paths for finding dependency and correlation between image pixels. Additionally, the densely connected convolutional blocks are kept in the final encoding layer to reuse features and prevent redundant data sharing. The method applied batch-normalization for reducing internal covariate shift in data distributions. The empirical evidence shows a promising response to our method compared with other semantic segmentation techniques.

Keywords: deep learning, semantic segmentation, image analysis, pixels connection, convolution neural network

Procedia PDF Downloads 102

Authors: Jimmy Eadie

Abstract:

Convolutional Cross Synthesis (CCS) has emerged as a powerful technique for blending input signals to create hybrid sounds. It has significantly expanded the horizons of digital signal processing, enabling artists to explore audio effects. However, the conventional applications of CCS primarily revolve around reverberation and room simulation rather than being utilized as a creative synthesis method. In this paper, we present the design of a digital instrument called CrossSampler that harnesses a parametric approach to convolution cross-synthesis, which involves using adjustable parameters to control the blending of audio signals through convolution. These parameters allow for customization of the resulting sound, offering greater creative control and flexibility. It enables users to shape the output by manipulating factors such as duration, intensity, and spectral characteristics. This approach facilitates experimentation and exploration in sound design and opens new sonic possibilities.

Keywords: convolution, synthesis, sampling, virtual instrument

Procedia PDF Downloads 63

Authors: Nirmal Yadav, Tanuja Srivastava

Abstract:

Shannon theory is an exact method to recover a band limited signals from its sampled values in discrete implementation, using sinc interpolators. But sinc based results are not much satisfactory for band-limited calculations so that convolution with window function, having compact support, has been introduced. Convolution Backprojection algorithm with window function is an approximation algorithm. In this paper, the error has been calculated, arises due to this approximation nature of reconstruction algorithm. This result will be defined for fan beam projection data which is more faster than parallel beam projection.

Keywords: computed tomography, convolution backprojection, radon transform, fan beam

Procedia PDF Downloads 490

Authors: Edward Holupka, John Rossman, Tye Morancy, Joseph Aronovitz, Irving Kaplan

Abstract:

A common modality for the treatment of early stage prostate cancer is the implantation of radioactive seeds directly into the prostate. The radioactive seeds are positioned inside the prostate to achieve optimal radiation dose coverage to the prostate. These radioactive seeds are positioned inside the prostate using Transrectal ultrasound imaging. Once all of the planned seeds have been implanted, two dimensional transaxial transrectal ultrasound images separated by 2 mm are obtained through out the prostate, beginning at the base of the prostate up to and including the apex. A common deep neural network, called DetectNet was trained to automatically determine the position of the implanted radioactive seeds within the prostate under ultrasound imaging. The results of the training using 950 training ultrasound images and 90 validation ultrasound images. The commonly used metrics for successful training were used to evaluate the efficacy and accuracy of the trained deep neural network and resulted in an loss_bbox (train) = 0.00, loss_coverage (train) = 1.89e-8, loss_bbox (validation) = 11.84, loss_coverage (validation) = 9.70, mAP (validation) = 66.87%, precision (validation) = 81.07%, and a recall (validation) = 82.29%, where train and validation refers to the training image set and validation refers to the validation training set. On the hardware platform used, the training expended 12.8 seconds per epoch. The network was trained for over 10,000 epochs. In addition, the seed locations as determined by the Deep Neural Network were compared to the seed locations as determined by a commercial software based on a one to three months after implant CT. The Deep Learning approach was within \strikeout off\uuline off\uwave off2.29\uuline default\uwave default mm of the seed locations determined by the commercial software. The Deep Learning approach to the determination of radioactive seed locations is robust, accurate, and fast and well within spatial agreement with the gold standard of CT determined seed coordinates.

Keywords: prostate, deep neural network, seed implant, ultrasound

Procedia PDF Downloads 197

Authors: Mohammed Al Sulaimani, Hamad Al Manhi

Abstract:

With the development of Remote Sensing technology, the resolution of optical Remote Sensing images has greatly improved, and images have become largely available. Numerous detectors have been developed for detecting different types of objects. In the past few years, Remote Sensing has benefited a lot from deep learning, particularly Deep Convolution Neural Networks (CNNs). Deep learning holds great promise to fulfill the challenging needs of Remote Sensing and solving various problems within different fields and applications. The use of Unmanned Aerial Systems in acquiring Aerial Photos has become highly used and preferred by most organizations to support their activities because of their high resolution and accuracy, which make the identification and detection of very small features much easier than Satellite Images. And this has opened an extreme era of Deep Learning in different applications not only in feature extraction and prediction but also in analysis. This work addresses the capacity of Machine Learning and Deep Learning in detecting and extracting Oil Leaks from Flowlines (Onshore) using High-Resolution Aerial Photos which have been acquired by UAS fixed with RGB Sensor to support early detection of these leaks and prevent the company from the leak’s losses and the most important thing environmental damage. Here, there are two different approaches and different methods of DL have been demonstrated. The first approach focuses on detecting the Oil Leaks from the RAW Aerial Photos (not processed) using a Deep Learning called Single Shoot Detector (SSD). The model draws bounding boxes around the leaks, and the results were extremely good. The second approach focuses on detecting the Oil Leaks from the Ortho-mosaiced Images (Georeferenced Images) by developing three Deep Learning Models using (MaskRCNN, U-Net and PSP-Net Classifier). Then, post-processing is performed to combine the results of these three Deep Learning Models to achieve a better detection result and improved accuracy. Although there is a relatively small amount of datasets available for training purposes, the Trained DL Models have shown good results in extracting the extent of the Oil Leaks and obtaining excellent and accurate detection.

Keywords: GIS, remote sensing, oil leak detection, machine learning, aerial photos, unmanned aerial systems

Procedia PDF Downloads 32

Authors: Konstantinos Perifanos, Eirini Florou, Dionysis Goutsos

Abstract:

This paper presents and benchmarks a number of end-to-end Deep Learning based models for metaphor detection in Greek. We combine Convolutional Neural Networks and Recurrent Neural Networks with representation learning to bear on the metaphor detection problem for the Greek language. The models presented achieve exceptional accuracy scores, significantly improving the previous state-of-the-art results, which had already achieved accuracy 0.82. Furthermore, no special preprocessing, feature engineering or linguistic knowledge is used in this work. The methods presented achieve accuracy of 0.92 and F-score 0.92 with Convolutional Neural Networks (CNNs) and bidirectional Long Short Term Memory networks (LSTMs). Comparable results of 0.91 accuracy and 0.91 F-score are also achieved with bidirectional Gated Recurrent Units (GRUs) and Convolutional Recurrent Neural Nets (CRNNs). The models are trained and evaluated only on the basis of training tuples, the related sentences and their labels. The outcome is a state-of-the-art collection of metaphor detection models, trained on limited labelled resources, which can be extended to other languages and similar tasks.

Keywords: metaphor detection, deep learning, representation learning, embeddings

Procedia PDF Downloads 152

Authors: Jiajun Wang, Xiaoge Li

Abstract:

The purpose of the aspect-level sentiment analysis task is to identify the sentiment polarity of aspects in a sentence. Currently, most methods mainly focus on using neural networks and attention mechanisms to model the relationship between aspects and context, but they ignore the dependence of words in different ranges in the sentence, resulting in deviation when assigning relationship weight to other words other than aspect words. To solve these problems, we propose a new aspect-level sentiment analysis model that combines a multi-channel convolutional network and graph convolutional network (GCN). Firstly, the context and the degree of association between words are characterized by Long Short-Term Memory (LSTM) and self-attention mechanism. Besides, a multi-channel convolutional network is used to extract the features of words in different ranges. Finally, a convolutional graph network is used to associate the node information of the dependency tree structure. We conduct experiments on four benchmark datasets. The experimental results are compared with those of other models, which shows that our model is better and more effective.

Keywords: aspect-level sentiment analysis, attention, multi-channel convolution network, graph convolution network, dependency tree

Procedia PDF Downloads 215

Authors: Jiahe Liu, HongyangYu, Feng Qian, Miao Luo

Abstract:

This study aimed to address the problem of improving the confidence of key points by fusing multi-view information, thereby estimating human posture more accurately. We first obtained multi-view image information and then used the MvP algorithm to fuse this multi-view information together to obtain a set of high-confidence human key points. We used these as the input for the Spatio-Temporal Graph Convolution (ST-GCN). ST-GCN is a deep learning model used for processing spatio-temporal data, which can effectively capture spatio-temporal relationships in video sequences. By using the MvP algorithm to fuse multi-view information and inputting it into the spatio-temporal graph convolution model, this study provides an effective method to improve the accuracy of human posture estimation and provides strong support for further research and application in related fields.

Keywords: multi-view, pose estimation, ST-GCN, joint fusion

Procedia PDF Downloads 69

Authors: T. Damak, O. Kriaa, A. Baccar, M. A. Ben Ayed, N. Masmoudi

Abstract:

In the last few years, Automatic Number Plate Recognition (ANPR) systems have become widely used in the safety, the security, and the commercial aspects. Forethought, several methods and techniques are computing to achieve the better levels in terms of accuracy and real time execution. This paper proposed a computer vision algorithm of Number Plate Localization (NPL) and Characters Segmentation (CS). In addition, it proposed an improved method in Optical Character Recognition (OCR) based on Deep Learning (DL) techniques. In order to identify the number of detected plate after NPL and CS steps, the Convolutional Neural Network (CNN) algorithm is proposed. A DL model is developed using four convolution layers, two layers of Maxpooling, and six layers of fully connected. The model was trained by number image database on the Jetson TX2 NVIDIA target. The accuracy result has achieved 95.84%.

Keywords: ANPR, CS, CNN, deep learning, NPL

Procedia PDF Downloads 305

Authors: Kaustubh Chakradeo, Michael Delves, Sofya Titarenko

Abstract:

Malaria is a serious disease which affects hundreds of millions of people around the world, each year. If not treated in time, it can be fatal. Despite recent developments in malaria diagnostics, the microscopy method to detect malaria remains the most common. Unfortunately, the accuracy of microscopic diagnostics is dependent on the skill of the microscopist and limits the throughput of malaria diagnosis. With the development of Artificial Intelligence tools and Deep Learning techniques in particular, it is possible to lower the cost, while achieving an overall higher accuracy. In this paper, we present a VGG-based model and compare it with previously developed models for identifying infected cells. Our model surpasses most previously developed models in a range of the accuracy metrics. The model has an advantage of being constructed from a relatively small number of layers. This reduces the computer resources and computational time. Moreover, we test our model on two types of datasets and argue that the currently developed deep-learning-based methods cannot efficiently distinguish between infected and contaminated cells. A more precise study of suspicious regions is required.

Keywords: convolution neural network, deep learning, malaria, thin blood smears

Procedia PDF Downloads 128

Authors: Arshia Aflaki, Hadis Karimipour, Anik Islam

Abstract:

Generative Adversarial Attacks (GAAs) threaten critical sectors, ranging from fingerprint recognition to industrial control systems. Existing Deep Learning (DL) algorithms are not robust enough against this kind of cyber-attack. As one of the most critical industries in the world, the power grid is not an exception. In this study, a Deep Reinforcement Learning-based (DRL) framework assisting the DL model to improve the robustness of the model against generative adversarial attacks is proposed. Real-world smart grid stability data, as an IIoT dataset, test our method and improves the classification accuracy of a deep learning model from around 57 percent to 96 percent.

Keywords: generative adversarial attack, deep reinforcement learning, deep learning, IIoT, generative adversarial networks, power system

Procedia PDF Downloads 35

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 157

Authors: Amin Nezarat, Naeime Seifadini

Abstract:

Trip mode selection is a behavioral characteristic of passengers with immense importance for travel demand analysis, transportation planning, and traffic management. Identification of trip mode distribution will allow transportation authorities to adopt appropriate strategies to reduce travel time, traffic and air pollution. The majority of existing trip mode inference models operate based on human selected features and traditional machine learning algorithms. However, human selected features are sensitive to changes in traffic and environmental conditions and susceptible to personal biases, which can make them inefficient. One way to overcome these problems is to use neural networks capable of extracting high-level features from raw input. In this study, the convolutional neural network (CNN) architecture is used to predict the trip mode distribution based on raw GPS trajectory data. The key innovation of this paper is the design of the layout of the input layer of CNN as well as normalization operation, in a way that is not only compatible with the CNN architecture but can also represent the fundamental features of motion including speed, acceleration, jerk, and Bearing rate. The highest prediction accuracy achieved with the proposed configuration for the convolutional neural network with batch normalization is 85.26%.

Keywords: predicting, deep learning, neural network, urban trip

Procedia PDF Downloads 137

Authors: Zhifeng Kong

Abstract:

Over-parameterized neural networks have attracted a great deal of attention in recent deep learning theory research, as they challenge the classic perspective of over-fitting when the model has excessive parameters and have gained empirical success in various settings. While a number of theoretical works have been presented to demystify properties of such models, the convergence properties of such models are still far from being thoroughly understood. In this work, we study the convergence properties of training two-hidden-layer partially over-parameterized fully connected networks with the Rectified Linear Unit activation via gradient descent. To our knowledge, this is the first theoretical work to understand convergence properties of deep over-parameterized networks without the equally-wide-hidden-layer assumption and other unrealistic assumptions. We provide a probabilistic lower bound of the widths of hidden layers and proved linear convergence rate of gradient descent. We also conducted experiments on synthetic and real-world datasets to validate our theory.

Keywords: over-parameterization, rectified linear units ReLU, convergence, gradient descent, neural networks

Procedia PDF Downloads 142

Authors: Amal Khalifa, Nicolas Vana Santos

Abstract:

Steganography has been known for centuries as an efficient approach for covert communication. Due to its popularity and ease of access, image steganography has attracted researchers to find secure techniques for hiding information within an innocent looking cover image. In this research, we propose a novel deep-learning approach to digital image steganography. The proposed method, DeepWaveletFusion, uses convolutional neural networks (CNN) to hide a secret image into a cover image of the same size. Two CNNs are trained back-to-back to merge the Discrete Wavelet Transform (DWT) of both colored images and eventually be able to blindly extract the hidden image. Based on two different image similarity metrics, a weighted gain function is used to guide the learning process and maximize the quality of the retrieved secret image and yet maintaining acceptable imperceptibility. Experimental results verified the high recoverability of DeepWaveletFusion which outperformed similar deep-learning-based methods.

Keywords: deep learning, steganography, image, discrete wavelet transform, fusion

Procedia PDF Downloads 89

Authors: I Gede Puja Astawa, Yoedy Moegiharto, Ahmad Zainudin, Imam Dui Agus Salim, Nur Annisa Anggraeni

Abstract:

Performance of Orthogonal Frequency Division Multiplexing (OFDM) system can be improved by adding channel coding (error correction code) to detect and correct the errors that occur during data transmission. One can use the convolution code. This paper presents performance of OFDM using Space Time Block Codes (STBC) diversity technique use QAM modulation with code rate 1/2. The evaluation is done by analyzing the value of Bit Error Rate (BER) vs. Energy per Bit to Noise Power Spectral Density Ratio (Eb/No). This scheme is conducted 256 sub-carrier which transmits Rayleigh multipath channel in OFDM system. To achieve a BER of 10-3 is required 30 dB SNR in SISO-OFDM scheme. For 2x2 MIMO-OFDM scheme requires 10 dB to achieve a BER of 10-3. For 4x4 MIMO-OFDM scheme requires 5 dB while adding convolution in a 4x4 MIMO-OFDM can improve performance up to 0 dB to achieve the same BER. This proves the existence of saving power by 3 dB of 4x4 MIMO-OFDM system without coding, power saving 7 dB of 2x2 MIMO-OFDM system without coding and significant power savings from SISO-OFDM system.

Keywords: convolution code, OFDM, MIMO, QAM, BER

Procedia PDF Downloads 387

Authors: Felipe A. Guth, Shane Ward, Kevin McDonnell

Abstract:

The assessment of disease levels in crop fields is an important and time-consuming task that generally relies on expert knowledge of trained individuals. Image classification in agriculture problems historically has been based on classical machine learning strategies that make use of hand-engineered features in the top of a classification algorithm. This approach tends to not produce results with high accuracy and generalization to the classes classified by the system when the nature of the elements has a significant variability. The advent of deep convolutional neural networks has revolutionized the field of machine learning, especially in computer vision tasks. These networks have great resourcefulness of learning and have been applied successfully to image classification and object detection tasks in the last years. The objective of this work was to propose a new method based on deep learning convolutional neural networks towards the task of disease level monitoring. Common RGB images of winter wheat were obtained during a growing season. Five categories of disease levels presence were produced, in collaboration with agronomists, for the algorithm classification. Disease level tasks performed by experts provided ground truth data for the disease score of the same winter wheat plots were RGB images were acquired. The system had an overall accuracy of 84% on the discrimination of the disease level classes.

Keywords: crop disease assessment, deep learning, precision agriculture, residual neural networks

Procedia PDF Downloads 331

Authors: Li Yu, Chenlu Jiao, Hongrun Xin, Yan Wang, Kaiyang Wang

Abstract:

In order to study the impact of various factors on the housing price, we propose to build different prediction models based on deep learning to determine the existing data of the real estate in order to more accurately predict the housing price or its changing trend in the future. Considering that the factors which affect the housing price vary widely, the proposed prediction models include two categories. The first one is based on multiple characteristic factors of the real estate. We built Convolution Neural Network (CNN) prediction model and Long Short-Term Memory (LSTM) neural network prediction model based on deep learning, and logical regression model was implemented to make a comparison between these three models. Another prediction model is time series model. Based on deep learning, we proposed an LSTM-1 model purely regard to time series, then implementing and comparing the LSTM model and the Auto-Regressive and Moving Average (ARMA) model. In this paper, comprehensive study of the second-hand housing price in Beijing has been conducted from three aspects: crawling and analyzing, housing price predicting, and the result comparing. Ultimately the best model program was produced, which is of great significance to evaluation and prediction of the housing price in the real estate industry.

Keywords: deep learning, convolutional neural network, LSTM, housing prediction

Procedia PDF Downloads 305

Authors: Thiago Spilborghs Bueno Meyer, Plinio Thomaz Aquino Junior

Abstract:

Through speech, which privileges the functional and interactive nature of the text, it is possible to ascertain the spatiotemporal circumstances, the conditions of production and reception of the discourse, the explicit purposes such as informing, explaining, convincing, etc. These conditions allow bringing the interaction between humans closer to the human-robot interaction, making it natural and sensitive to information. However, it is not enough to understand what is said; it is necessary to recognize emotions for the desired interaction. The validity of the use of neural networks for feature selection and emotion recognition was verified. For this purpose, it is proposed the use of neural networks and comparison of models, such as recurrent neural networks and deep neural networks, in order to carry out the classification of emotions through speech signals to verify the quality of recognition. It is expected to enable the implementation of robots in a domestic environment, such as the HERA robot from the RoboFEI@Home team, which focuses on autonomous service robots for the domestic environment. Tests were performed using only the Mel-Frequency Cepstral Coefficients, as well as tests with several characteristics of Delta-MFCC, spectral contrast, and the Mel spectrogram. To carry out the training, validation and testing of the neural networks, the eNTERFACE’05 database was used, which has 42 speakers from 14 different nationalities speaking the English language. The data from the chosen database are videos that, for use in neural networks, were converted into audios. It was found as a result, a classification of 51,969% of correct answers when using the deep neural network, when the use of the recurrent neural network was verified, with the classification with accuracy equal to 44.09%. The results are more accurate when only the Mel-Frequency Cepstral Coefficients are used for the classification, using the classifier with the deep neural network, and in only one case, it is possible to observe a greater accuracy by the recurrent neural network, which occurs in the use of various features and setting 73 for batch size and 100 training epochs.

Keywords: emotion recognition, speech, deep learning, human-robot interaction, neural networks

Procedia PDF Downloads 169