scholarly journals Partial-Gated Memristor Crossbar for Fast and Power-Efficient Defect-Tolerant Training

Micromachines ◽  
2019 ◽  
Vol 10 (4) ◽  
pp. 245 ◽  
Author(s):  
Pham ◽  
Nguyen ◽  
Min
Keyword(s):  
Circuit Simulation ◽  
Recognition Rate ◽  
Back Propagation ◽  
Fine Tuning ◽  
Back Propagation Algorithm ◽  
Power Efficient ◽  
Long Time ◽  
Memristor Crossbar ◽  
Programming Time ◽  
Gating Scheme

A real memristor crossbar has defects, which should be considered during the retraining time after the pre-training of the crossbar. For retraining the crossbar with defects, memristors should be updated with the weights that are calculated by the back-propagation algorithm. Unfortunately, programming the memristors takes a very long time and consumes a large amount of power, because of the incremental behavior of memristor’s program-verify scheme for the fine-tuning of memristor’s conductance. To reduce the programming time and power, the partial gating scheme is proposed here to realize the partial training, where only some part of neurons are trained, which are more responsible in the recognition error. By retraining the part, rather than the entire crossbar, the programming time and power of memristor crossbar can be significantly reduced. The proposed scheme has been verified by CADENCE circuit simulation with the real memristor’s Verilog-A model. When compared to retraining the entire crossbar, the loss of recognition rate of the partial gating scheme has been estimated only as small as 2.5% and 2.9%, for the MNIST and CIFAR-10 datasets, respectively. However, the programming time and power can be saved by 86% and 89.5% than the 100% retraining, respectively.

Construction of a deep sparse filtering network for rotating machinery fault diagnosis

2021 ◽  
pp. 095440702110148
Author(s):  
Chun Cheng ◽  
Wei Zou ◽  
Weiping Wang ◽  
Michael Pecht
Keyword(s):  
Fault Diagnosis ◽  
Back Propagation ◽  
Feature Learning ◽  
Back Propagation Neural Network ◽  
Rotating Machinery ◽  
Fine Tuning ◽  
Local Optimum ◽  
Back Propagation Algorithm ◽  
Intelligent Fault Diagnosis ◽  
Sparse Filtering

Deep neural networks (DNNs) have shown potential in intelligent fault diagnosis of rotating machinery. However, traditional DNNs such as the back-propagation neural network are highly sensitive to the initial weights and easily fall into the local optimum, which restricts the feature learning capability and diagnostic performance. To overcome the above problems, a deep sparse filtering network (DSFN) constructed by stacked sparse filtering is developed in this paper and applied to fault diagnosis. The developed DSFN is pre-trained by sparse filtering in an unsupervised way. The back-propagation algorithm is employed to optimize the DSFN after pre-training. Then, the DSFN-based intelligent fault diagnosis method is validated using two experiments. The results show that pre-training with sparse filtering and fine-tuning can help the DSFN search for the optimal network parameters, and the DSFN can learn discriminative features adaptively from rotating machinery datasets. Compared with classical methods, the developed diagnostic method can diagnose rotating machinery faults with higher accuracy using fewer training samples.

A Model with Back-Propagation Algorithm for the Estimation of Irreducible Water Saturation

2011 ◽  
Vol 361-363 ◽  
pp. 445-450 ◽  
Author(s):  
Ping Hua Ma ◽  
Hong Fu Fan ◽  
Ke Li
Keyword(s):  
Prediction Models ◽  
Water Saturation ◽  
Back Propagation ◽  
Superior Performance ◽  
Back Propagation Algorithm ◽  
Irreducible Water Saturation ◽  
Long Time ◽  
Multi Phase Flow ◽  
Conventional Models ◽  
Hidden Layer

As one of the most important reservoir parameters, irreducible water saturation, Swir, is a key parameter in evaluating multi-phase flow, as well as its importance in defining oil in-place. Residual oil saturation, the target of tertiary recovery, is also a function of Swir. In traditionally, Swir is determined by conducting capillary pressure experiments, requiring considerable resources and long time periods, with the consequence of a limited number of core plug evaluations for a particular reservoir. Thus, the estimation of Swir with mathematical models is developed in recent years. The study reported in this paper uses artificial neural network to determine Swir. The optimal model is chosen among 25 simulations, subtilizing different combinations of hidden layer nodes and activation functions for the hidden and output layers. Its performance is compared with other conventional models, demonstrating the superior performance of the proposed Swir prediction models.

A Model with Back-Propagation Algorithm for Recovery Efficiency Forecast of Carbon Dioxide Flooding

2013 ◽  
Vol 295-298 ◽  
pp. 3137-3141
Author(s):  
Ping Hua Ma ◽  
Hong Fu Fan ◽  
Ke Li
Keyword(s):  
Back Propagation ◽  
Superior Performance ◽  
Recovery Efficiency ◽  
Co2 Flooding ◽  
Back Propagation Algorithm ◽  
Optimal Model ◽  
Long Time ◽  
Tertiary Recovery ◽  
Hidden Layer ◽  
Key Parameter

As one of the most important displacements in producing petroleum, CO2 flooding has been developed for nearly 50 years around the world in order to improve the tertiary recovery. The recovery efficiency, R, is a key parameter in the CO2 displacement of crude oil. Traditionally, R is determined by conducting CO2 flooding experiment, requiring considerable resources and long time periods, with the consequence of a limited number of core plug evaluations for a particular reservoir. Thus, the estimation of R with mathematical models is developed in recent years, which also needs plenty of relevant parameters considered. The study reported in this paper uses artificial neural network to determine R. Five dimensionless variables are considered to analyse the CO2 immiscible displacement process. An optimal model is chosen with its suitable hidden layer nodes and activation functions for the hidden and output layers. Its performance is compared with the numerical simulation model, demonstrating the superior performance of the proposed R prediction model.

scholarly journals CONTROLLING OF AN INDUSTRIAL ROBOTIC ARM

2013 ◽  
pp. 230-235
Author(s):  
V. VENKATESWARLU ◽  
S. SREENIVASULU
Keyword(s):  
Response Time ◽  
Industrial Robot ◽  
Low Cost ◽  
Recognition Rate ◽  
Back Propagation ◽  
Industrial Robots ◽  
System Response ◽  
Back Propagation Algorithm ◽  
System Response Time ◽  
Artificial Neural Network Ann

Most of industrial robots are still programmed using the typical teaching process, through the use of the robot teach pendant. In this paper is proposed an accelerometer-based system to control an industrial robot using two low-cost and small 3-axis wireless accelerometers. These accelerometers are attached to the human arms, capturing its behavior (gestures and postures). An Artificial Neural Network (ANN) trained with a back-propagation algorithm was used to recognize arm gestures and postures, which then will be used as input in the control of the robot. The aim is that the robot starts the movement almost at the same time as the user starts to perform a gesture or posture (low response time). The results show that the system allows the control of an industrial robot in an intuitive way. However, the achieved recognition rate of gestures and postures (92%) should be improved in future, keeping the compromise with the system response time (160 milliseconds). Finally, the results of some tests performed with an industrial robot are presented and discussed.

scholarly journals Robust features extraction for general fish classification

2019 ◽  
Vol 9 (6) ◽  
pp. 5192 ◽  
Author(s):  
Mutasem K. Alsmadi ◽  
Mohammed Tayfour ◽  
Raed A. Alkhasawneh ◽  
Usama Badawi ◽  
Ibrahim Almarashdeh ◽  
...  
Keyword(s):  
Memetic Algorithm ◽  
Recognition Rate ◽  
Back Propagation ◽  
Experimental Tests ◽  
Back Propagation Algorithm ◽  
Fish Family ◽  
Propagation Algorithm ◽  
Combination Theory ◽  
Anchor Points ◽  
Classification Evaluation

Image recognition process could be plagued by many problems including noise, overlap, distortion, errors in the outcomes of segmentation, and impediment of objects within the image. Based on feature selection and combination theory between major extracted features, this study attempts to establish a system that could recognize fish object within the image utilizing texture, anchor points, and statistical measurements. Then, a generic fish classification is executed with the application of an innovative classification evaluation through a meta-heuristic algorithm known as Memetic Algorithm (Genetic Algorithm with Simulated Annealing) with back-propagation algorithm (MA-B Classifier). Here, images of dangerous and non-dangerous fish are recognized. Images of dangerous fish are further recognized as Predatory or Poison fish family, whereas families of non-dangerous fish are classified into garden and food family.  A total of 24 fish families were used in testing the proposed prototype, whereby each family encompasses different number of species. The process of classification was successfully undertaken by the proposed prototype, whereby 400 distinct fish images were used in the experimental tests. Of these fish images, 250 were used for training phase while 150 were used for testing phase. The back-propagation algorithm and the proposed MA-B Classifier produced a general accuracy recognition rate of 82.25 and 90% respectively.

Deep Learning Approach for Devanagari Script Recognition

2017 ◽  
Vol 17 (03) ◽  
pp. 1750016 ◽  
Author(s):  
S. Prabhanjan ◽  
R. Dinesh
Keyword(s):  
Deep Learning ◽  
Classification System ◽  
Recognition Performance ◽  
Recognition Rate ◽  
Back Propagation ◽  
Feature Representation ◽  
Fine Tuning ◽  
Large Set ◽  
Restricted Boltzmann Machines ◽  
Good Feature

In this paper, we have proposed a new technique for recognition of handwritten Devanagari Script using deep learning architecture. In any OCR or classification system extracting discriminating feature is most important and crucial step for its success. Accuracy of such system often depends on the good feature representation. Deciding upon the appropriate features for classification system is highly subjective and requires lot of experience to decide proper set of features for a given classification system. For handwritten Devanagari characters it is very difficult to decide on optimal set of good feature to get good recognition rate. These methods use raw pixel values as features. Deep Learning architectures learn hierarchies of features. In this work, first image is preprocessed to remove noise, converted to binary image, resized to fixed size of 30[Formula: see text][Formula: see text][Formula: see text]40 and then convert to gray scale image using mask operation, it blurs the edges of the images. Then we learn features using an unsupervised stacked Restricted Boltzmann Machines (RBM) and use it with the deep belief network for recognition. Finally network weight parameters are fine tuned by supervised back propagation learning to improve the overall recognition performance. The proposed method has been tested on large set of handwritten numerical, character, vowel modifiers and compound characters and experimental results reveals that unsupervised method yields very good accuracy of (83.44%) and upon fine tuning of network parameters with supervised learning yields accuracy of (91.81%) which is the best results reported so far for handwritten Devanagari characters.

scholarly journals Two-Stage Water Jet Landing Point Prediction Model for Intelligent Water Shooting Robot

Sensors ◽  
2021 ◽  
Vol 21 (8) ◽  
pp. 2704
Author(s):  
Yunhan Lin ◽  
Wenlong Ji ◽  
Haowei He ◽  
Yaojie Chen
Keyword(s):  
Pitch Angle ◽  
Back Propagation ◽  
Water Jet ◽  
The State ◽  
Model Parameters ◽  
Propagation Time ◽  
Back Propagation Algorithm ◽  
Angle Error ◽  
Landing Point ◽  
Propagation Algorithm

In this paper, an intelligent water shooting robot system for situations of carrier shake and target movement is designed, which uses a 2 DOF (degree of freedom) robot as an actuator, a photoelectric camera to detect and track the desired target, and a gyroscope to keep the robot’s body stable when it is mounted on the motion carriers. Particularly, for the accurate shooting of the designed system, an online tuning model of the water jet landing point based on the back-propagation algorithm was proposed. The model has two stages. In the first stage, the polyfit function of Matlab is used to fit a model that satisfies the law of jet motion in ideal conditions without interference. In the second stage, the model uses the back-propagation algorithm to update the parameters online according to the visual feedback of the landing point position. The model established by this method can dynamically eliminate the interference of external factors and realize precise on-target shooting. The simulation results show that the model can dynamically adjust the parameters according to the state relationship between the landing point and the desired target, which keeps the predicted pitch angle error within 0.1°. In the test on the actual platform, when the landing point is 0.5 m away from the position of the desired target, the model only needs 0.3 s to adjust the water jet to hit the target. Compared to the state-of-the-art method, GA-BP (genetic algorithm-back-propagation), the proposed method’s predicted pitch angle error is within 0.1 degree with 1/4 model parameters, while costing 1/7 forward propagation time and 1/200 back-propagation calculation time.

scholarly journals Deep neural networks using a single neuron: folded-in-time architecture using feedback-modulated delay loops

2021 ◽  
Vol 12 (1) ◽  
Author(s):  
Florian Stelzer ◽  
André Röhm ◽  
Raul Vicente ◽  
Ingo Fischer ◽  
Serhiy Yanchuk
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Deep Neural Network ◽  
Single Neuron ◽  
Deep Neural Networks ◽  
Back Propagation ◽  
Local Network ◽  
Multiple Time ◽  
Learning Tools ◽  
Back Propagation Algorithm

AbstractDeep neural networks are among the most widely applied machine learning tools showing outstanding performance in a broad range of tasks. We present a method for folding a deep neural network of arbitrary size into a single neuron with multiple time-delayed feedback loops. This single-neuron deep neural network comprises only a single nonlinearity and appropriately adjusted modulations of the feedback signals. The network states emerge in time as a temporal unfolding of the neuron’s dynamics. By adjusting the feedback-modulation within the loops, we adapt the network’s connection weights. These connection weights are determined via a back-propagation algorithm, where both the delay-induced and local network connections must be taken into account. Our approach can fully represent standard Deep Neural Networks (DNN), encompasses sparse DNNs, and extends the DNN concept toward dynamical systems implementations. The new method, which we call Folded-in-time DNN (Fit-DNN), exhibits promising performance in a set of benchmark tasks.

Sign in / Sign up

Export Citation Format

Share Document