Abstract: Acoustic sensors, if embedded in smart street lights, can help in capturing the activities (car honking, sirens, events, traffic, etc.) in cities. Needless to say, the acoustic data from such scenarios are complex due to multiple audio streams originating from different events, and when decomposed to independent signals, the amount of retrieved data volume is small in quantity which is inadequate to train deep neural networks. So, in this paper, we address the two challenges: a) separating the mixed signals, and b) developing an efficient acoustic classifier under data paucity. So, to address these challenges, we propose an architecture with supervised deep learning, where the initial captured mixed acoustics data are analyzed with Fast Fourier Transformation (FFT), followed by filtering the noise from the signal, and then decomposed to independent signals by fast independent component analysis (Fast ICA). To address the challenge of data paucity, we propose a multi feature-based deep neural network with high performance that is reflected in our experiments when compared to the conventional convolutional neural network (CNN) and multi-layer perceptron (MLP).
Abstract: The Earth's surface is constantly changing through forces of nature and human activities. Reliable, accurate, and timely change detection is critical to environmental monitoring, resource management, and planning activities. Recently, interest in deep learning algorithms, especially convolutional neural networks, has increased in the field of image change detection due to their powerful ability to extract multi-level image features automatically. However, these networks are prone to drawbacks that limit their applications, which reside in their inability to capture spatial relationships between image instances, as this necessitates a large amount of training data. As an alternative, Capsule Network has been proposed to overcome these shortcomings. Although its effectiveness in remote sensing image analysis has been experimentally verified, its application in change detection tasks remains very sparse. Motivated by its greater robustness towards improved hierarchical object representation, this study aims to apply a capsule network for PolSAR image Change Detection. The experimental results demonstrate that the proposed change detection method can yield a significantly higher detection rate compared to methods based on convolutional neural networks.
Abstract: With the rapid development of deep learning, neural network and deep learning algorithms play a significant role in various practical applications. Due to the high accuracy and good performance, Convolutional Neural Networks (CNNs) especially have become a research hot spot in the past few years. However, the size of the networks becomes increasingly large scale due to the demands of the practical applications, which poses a significant challenge to construct a high-performance implementation of deep learning neural networks. Meanwhile, many of these application scenarios also have strict requirements on the performance and low-power consumption of hardware devices. Therefore, it is particularly critical to choose a moderate computing platform for hardware acceleration of CNNs. This article aimed to survey the recent advance in Field Programmable Gate Array (FPGA)-based acceleration of CNNs. Various designs and implementations of the accelerator based on FPGA under different devices and network models are overviewed, and the versions of Graphic Processing Units (GPUs), Application Specific Integrated Circuits (ASICs) and Digital Signal Processors (DSPs) are compared to present our own critical analysis and comments. Finally, we give a discussion on different perspectives of these acceleration and optimization methods on FPGA platforms to further explore the opportunities and challenges for future research. More helpfully, we give a prospect for future development of the FPGA-based accelerator.
Abstract: The reason for conducting this research is to develop an algorithm that is capable of classifying news articles from the automobile industry, according to the competitive actions that they entail, with the use of Text Mining (TM) methods. It is needed to test how to properly preprocess the data for this research by preparing pipelines which fits each algorithm the best. The pipelines are tested along with nine different classification algorithms in the realm of regression, support vector machines, and neural networks. Preliminary testing for identifying the optimal pipelines and algorithms resulted in the selection of two algorithms with two different pipelines. The two algorithms are Logistic Regression (LR) and Artificial Neural Network (ANN). These algorithms are optimized further, where several parameters of each algorithm are tested. The best result is achieved with the ANN. The final model yields an accuracy of 0.79, a precision of 0.80, a recall of 0.78, and an F1 score of 0.76. By removing three of the classes that created noise, the final algorithm is capable of reaching an accuracy of 94%.
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.
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.
Abstract: Due to the sensor technology, video surveillance has become the main way for security control in every big city in the world. Surveillance is usually used by governments for intelligence gathering, the prevention of crime, the protection of a process, person, group or object, or the investigation of crime. Many surveillance systems based on computer vision technology have been developed in recent years. Moving target tracking is the most common task for Unmanned Aerial Vehicle (UAV) to find and track objects of interest in mobile aerial surveillance for civilian applications. The paper is focused on vision-based collision avoidance for UAVs by recurrent neural networks. First, images from cameras on UAV were fused based on deep convolutional neural network. Then, a recurrent neural network was constructed to obtain high-level image features for object tracking and extracting low-level image features for noise reducing. The system distributed the calculation of the whole system to local and cloud platform to efficiently perform object detection, tracking and collision avoidance based on multiple UAVs. The experiments on several challenging datasets showed that the proposed algorithm outperforms the state-of-the-art methods.
Abstract: This paper presents a prediction performance of
feedforward Multilayer Perceptron (MLP) and Echo State Networks
(ESN) trained with extended Kalman filter. Feedforward neural
networks and ESN are powerful neural networks which can track and
predict nonlinear signals. However, their tracking performance
depends on the specific signals or data sets, having the risk of
instability accompanied by large error. In this study we explore this
process by applying different network size and leaking rate for
prediction of nonlinear or chaotic signals in MLP neural networks.
Major problems of ESN training such as the problem of initialization
of the network and improvement in the prediction performance are
tackled. The influence of coefficient of activation function in the
hidden layer and other key parameters are investigated by simulation
results. Extended Kalman filter is employed in order to improve the
sequential and regulation learning rate of the feedforward neural
networks. This training approach has vital features in the training of
the network when signals have chaotic or non-stationary sequential
pattern. Minimization of the variance in each step of the computation
and hence smoothing of tracking were obtained by examining the
results, indicating satisfactory tracking characteristics for certain
conditions. In addition, simulation results confirmed satisfactory
performance of both of the two neural networks with modified
parameterization in tracking of the nonlinear signals.
Abstract: This paper presents a deep-learning mechanism for classifying computer generated images and photographic images. The proposed method accounts for a convolutional layer capable of automatically learning correlation between neighbouring pixels. In the current form, Convolutional Neural Network (CNN) will learn features based on an image's content instead of the structural features of the image. The layer is particularly designed to subdue an image's content and robustly learn the sensor pattern noise features (usually inherited from image processing in a camera) as well as the statistical properties of images. The paper was assessed on latest natural and computer generated images, and it was concluded that it performs better than the current state of the art methods.
Abstract: Green microgrids using mostly renewable energy (RE) for generation, are complex systems with inherent nonlinear dynamics. Among a variety of different optimization tools there are only a few ones that adequately consider this complexity. This paper evaluates applicability of two somewhat similar optimization tools tailored for standalone RE microgrids and also assesses a machine learning tool for performance prediction that can enhance the reliability of any chosen optimization tool. It shows that one of these microgrid optimization tools has certain advantages over another and presents a detailed routine of preparing input data to simulate RE microgrid behavior. The paper also shows how neural-network-based predictive modeling can be used to validate and forecast solar power generation based on weather time series data, which improves the overall quality of standalone RE microgrid analysis.
Abstract: Crude oil market is an immensely complex and dynamic environment and thus the task of predicting changes in such an environment becomes challenging with regards to its accuracy. A number of approaches have been adopted to take on that challenge and machine learning has been at the core in many of them. There are plenty of examples of algorithms based on machine learning yielding satisfactory results for such type of prediction. In this paper, we have tried to predict crude oil prices using Long Short-Term Memory (LSTM) based recurrent neural networks. We have tried to experiment with different types of models using different epochs, lookbacks and other tuning methods. The results obtained are promising and presented a reasonably accurate prediction for the price of crude oil in near future.
Abstract: In this paper, we want to determine whether the accurate prediction of nicotine concentration can be obtained by using a network of smart objects and an e-cigarette. The approach consists of, first, the recognition of factors influencing smoking cessation such as physical activity recognition and participant’s behaviors (using both smartphone and smartwatch), then the prediction of the configuration of the e-cigarette (in terms of nicotine concentration, power, and resistance of e-cigarette). The study uses a network of commonly connected objects; a smartwatch, a smartphone, and an e-cigarette transported by the participants during an uncontrolled experiment. The data obtained from sensors carried in the three devices were trained by a Long short-term memory algorithm (LSTM). Results show that our LSTM-based model allows predicting the configuration of the e-cigarette in terms of nicotine concentration, power, and resistance with a root mean square error percentage of 12.9%, 9.15%, and 11.84%, respectively. This study can help to better control consumption of nicotine and offer an intelligent configuration of the e-cigarette to users.
Abstract: This paper addresses the issues and technique for textile industry using data mining techniques. Data mining has been applied to the stitching of garments products that were obtained from a textile company. Data mining techniques were applied to the data obtained from the CHAID algorithm, CART algorithm, Regression Analysis and, Artificial Neural Networks. Classification technique based analyses were used while data mining and decision model about the production per person and variables affecting about production were found by this method. In the study, the results show that as the daily working time increases, the production per person also decreases. In addition, the relationship between total daily working and production per person shows a negative result and the production per person show the highest and negative relationship.
Abstract: Software maintenance phase is started once a software project has been developed and delivered. After that, any modification to it corresponds to maintenance. Software maintenance involves modifications to keep a software project usable in a changed or a changing environment, to correct discovered faults, and modifications, and to improve performance or maintainability. Software maintenance and management of software maintenance are recognized as two most important and most expensive processes in a life of a software product. This research is basing the prediction of maintenance, on risks and time evaluation, and using them as data sets for working with neural networks. The aim of this paper is to provide support to project maintenance managers. They will be able to pass the issues planned for the next software-service-patch to the experts, for risk and working time evaluation, and afterward to put all data to neural networks in order to get software maintenance prediction. This process will lead to the more accurate prediction of the working hours needed for the software-service-patch, which will eventually lead to better planning of budget for the software maintenance projects.
Abstract: The use of decision support systems in agriculture may help monitoring large fields of crops by automatically detecting the symptoms of foliage diseases. In our work, we designed and implemented a decision support system for small tomatoes producers. This work investigates ways to recognize the late blight disease from the analysis of digital images of tomatoes, using a pair of multilayer perceptron neural networks. The networks outputs are used to generate repainted tomato images in which the injuries on the plant are highlighted, and to calculate the damage level of each plant. Those levels are then used to construct a situation map of a farm where a cellular automata simulates the outbreak evolution over the fields. The simulator can test different pesticides actions, helping in the decision on when to start the spraying and in the analysis of losses and gains of each choice of action.
Abstract: The stereo matching problem, while having been present for several decades, continues to be an active area of research. The goal of this research is to find correspondences between elements found in a set of stereoscopic images. With these pairings, it is possible to infer the distance of objects within a scene, relative to the observer. Advancements in this field have led to experimentations with various techniques, from graph-cut energy minimization to artificial neural networks. At the basis of these techniques is a cost function, which is used to evaluate the likelihood of a particular match between points in each image. While at its core, the cost is based on comparing the image pixel data; there is a general lack of consistency as to what image data representation to use. This paper presents an experimental analysis to compare the effectiveness of more common image data representations. The goal is to determine the effectiveness of these data representations to reduce the cost for the correct correspondence relative to other possible matches.
Abstract: This paper presents the development of an algorithm that predicts the arrival of a secondary user (SU) to a base station (BS) in a cognitive network based on infrastructure, requesting a Best Effort (BE) or Real Time (RT) type of service with a determined bandwidth (BW) implementing neural networks. The algorithm dynamically uses a neural network construction technique using the geometric pyramid topology and trains a Multilayer Perceptron Neural Networks (MLPNN) based on the historical arrival of an SU to estimate future applications. This will allow efficiently managing the information in the BS, since it precedes the arrival of the SUs in the stage of selection of the best channel in CRN. As a result, the software application determines the probability of arrival at a future time point and calculates the performance metrics to measure the effectiveness of the predictions made.
Abstract: The complexity of scavenging by ports and its impact on engine efficiency create the need to understand and to model it as realistically as possible. However, there are few empirical scavenging models and these are highly specialized. In a design optimization process, they appear very restricted and their field of use is limited. This paper presents a comparison of two methods to establish and reduce a model of the scavenging process in 2-stroke diesel engines. To solve the lack of scavenging models, a CFD model has been developed and is used as the referent case. However, its large size requires a reduction. Two techniques have been tested depending on their fields of application: The NTF method and neural networks. They both appear highly appropriate drastically reducing the model’s size (over 90% reduction) with a low relative error rate (under 10%). Furthermore, each method produces a reduced model which can be used in distinct specialized fields of application: the distribution of a quantity (mass fraction for example) in the cylinder at each time step (pseudo-dynamic model) or the qualification of scavenging at the end of the process (pseudo-static model).
Abstract: Ionic liquids are finding a wide range of applications from reaction media to separations and materials processing. In these applications, Vapor–Liquid equilibrium (VLE) is the most important one. VLE for six systems at 353 K and activity coefficients at infinite dilution [(γ)_i^∞] for various solutes (alkanes, alkenes, cycloalkanes, cycloalkenes, aromatics, alcohols, ketones, esters, ethers, and water) in the ionic liquids (1-ethyl-3-methylimidazolium bis (trifluoromethylsulfonyl)imide [EMIM][BTI], 1-hexyl-3-methyl imidazolium bis (trifluoromethylsulfonyl) imide [HMIM][BTI], 1-octyl-3-methylimidazolium bis(trifluoromethylsulfonyl) imide [OMIM][BTI], and 1-butyl-1-methylpyrrolidinium bis (trifluoromethylsulfonyl) imide [BMPYR][BTI]) have been used to train neural networks in the temperature range from (303 to 333) K. Densities of the ionic liquids, Hildebrant constant of substances, and temperature were selected as input of neural networks. The networks with different hidden layers were examined. Networks with seven neurons in one hidden layer have minimum error and good agreement with experimental data.
Abstract: Mumbai, being traditionally the epicenter of India's
trade and commerce, the existing major ports such as Mumbai and
Jawaharlal Nehru Ports (JN) situated in Thane estuary are also
developing its waterfront facilities. Various developments over the
passage of decades in this region have changed the tidal flux
entering/leaving the estuary. The intake at Pir-Pau is facing the
problem of shortage of water in view of advancement of shoreline,
while jetty near Ulwe faces the problem of ship scheduling due to
existence of shallower depths between JN Port and Ulwe Bunder. In
order to solve these problems, it is inevitable to have information
about tide levels over a long duration by field measurements.
However, field measurement is a tedious and costly affair;
application of artificial intelligence was used to predict water levels
by training the network for the measured tide data for one lunar tidal
cycle. The application of two layered feed forward Artificial Neural
Network (ANN) with back-propagation training algorithms such as
Gradient Descent (GD) and Levenberg-Marquardt (LM) was used to
predict the yearly tide levels at waterfront structures namely at Ulwe
Bunder and Pir-Pau. The tide data collected at Apollo Bunder, Ulwe,
and Vashi for a period of lunar tidal cycle (2013) was used to train,
validate and test the neural networks. These trained networks having
high co-relation coefficients (R= 0.998) were used to predict the tide
at Ulwe, and Vashi for its verification with the measured tide for the
year 2000 & 2013. The results indicate that the predicted tide levels
by ANN give reasonably accurate estimation of tide. Hence, the
trained network is used to predict the yearly tide data (2015) for
Ulwe. Subsequently, the yearly tide data (2015) at Pir-Pau was
predicted by using the neural network which was trained with the
help of measured tide data (2000) of Apollo and Pir-Pau. The analysis of measured data and study reveals that: The
measured tidal data at Pir-Pau, Vashi and Ulwe indicate that there is
maximum amplification of tide by about 10-20 cm with a phase lag
of 10-20 minutes with reference to the tide at Apollo Bunder
(Mumbai). LM training algorithm is faster than GD and with increase
in number of neurons in hidden layer and the performance of the
network increases. The predicted tide levels by ANN at Pir-Pau and
Ulwe provides valuable information about the occurrence of high and
low water levels to plan the operation of pumping at Pir-Pau and
improve ship schedule at Ulwe.