scispace - formally typeset
Search or ask a question
Topic

Hybrid neural network

About: Hybrid neural network is a research topic. Over the lifetime, 1305 publications have been published within this topic receiving 18223 citations.


Papers
More filters
Proceedings ArticleDOI
12 Nov 2012
TL;DR: An unsupervised spike sorting method based on the hybrid neural network with principal component analysis network (PCAN) and normal boundary response (NBR) and self-organizing map network (SOMN) classifier to further improve the efficient and adaptive of classification system is proposed.
Abstract: Automatic efficient spike sorting is one of the biggest challenges for the neural recording microsystem online. An unsupervised spike sorting method is proposed in this paper, based on the hybrid neural network with principal component analysis network (PCAN) and normal boundary response (NBR) self-organizing map network (SOMN) classifier. The PCAN extracted the spike features with the dimension reduced and correlation eliminated; The SOM network perform the spike distribution in the feature space, thus after convergence, the weights of the neurons demonstrate the spike cluster distribution in the feature space; At last the spike sorting was finished by computing the neurons' Normal Boundary Response (NBR) which determined the neurons' classes. The experimental results show that, based on hybrid neural network spiking sorting algorithm, it can achieve the accuracy above 97.91% with signals containing five classes. The novel classification algorithm proposed is to further improve the efficient and adaptive of classification system.

2 citations

Book ChapterDOI
01 Jan 2013
TL;DR: In this paper, the use of a modified Bayesian back-propagation neural network to model the highly nonlinear relationships between interacting factors or variables has been explored for geotechnical engineering design.
Abstract: Empirical relationships are commonly used in geotechnical engineering design to estimate engineering properties of soils and evaluate the performance of geotechnical structures because the interacting factors and relationship between these factors are not precisely known. This chapter explores the use of a modified Bayesian back-propagation neural network to model the highly nonlinear relationships between interacting factors or variables. The main advantage of neural networks over conventional regression analysis techniques is that the neural network is able to find a best-fit solution without the need to specify the relationship or the form of the relationship between variables. It is, therefore, useful for analyzing problems where there is incomplete understanding of the problem to be solved, but where training data are available, as is the case for many geotechnical engineering problems. With the integration of the Bayesian framework into the back-propagation algorithm, error bars can be calculated for network output instead of just a single output value given in conventional back-propagation neural networks. These error bars indicate the confidence level of the predicted values in relation to the spatial density of the training data. This chapter describes the use of a hybrid neural network that incorporates the genetic algorithm search engine and Bayesian approach in quantifying the uncertainty of the learned model. Some practical examples of its application to pile foundation and retaining wall design are presented to demonstrate the usefulness of this hybrid model.

2 citations

Posted ContentDOI
04 May 2021-bioRxiv
TL;DR: In this paper, a hybrid neural network (HNN) was used to predict chemical carcinogenicity of a wide variety of real-life exposure chemicals in large scale, and three types of machine learning models were developed: binary classification models, multiclass classification models and regression models.
Abstract: Determining environmental chemical carcinogenicity is an urgent need as humans are increasingly exposed to these chemicals. In this study, we determined the carcinogenicity of wide variety real-life exposure chemicals in large scale. To determine chemical carcinogenicity, we have developed carcinogenicity prediction models based on the hybrid neural network (HNN) architecture. In the HNN model, we included new SMILES feature representation method, by modifying our previous 3D array representation of 1D SMILES simulated by the convolutional neural network (CNN). We used 653 molecular descriptors modeled by feed forward neural network (FFNN), and SMILES as chemical features to train the models. We have developed three types of machine learning models: binary classification models to predict chemical is a carcinogenic or non-carcinogenic, multiclass classification models to predict severity of the chemical carcinogenicity, and regression models to predict median toxic dose of the chemicals. Along with the hybrid neural network (HNN) model that we developed, Random Forest (RF), Bootstrap Aggregating (Bagging) and Adaptive Boosting (AdaBoost) methods were also used for binary and multiclass classification. Regression models were developed using HNN, RF, Support Vector Regressor (SVR), Gradient Boosting (GB), Kernel Ridge (KR), Decision Tree with AdaBoost (DT), KNeighbors (KN), and a consensus method. For binary classification, our HNN model predicted with an average accuracy of 74.33% and an average AUC of 0.806, for multiclass classification, the HNN model predicted with an average accuracy of 50.58% and an average micro-AUC of 0.68, and for regression model, the consensus method achieved R2 of 0.40. The predictive performance of our models based on a highly diverse chemicals is comparable to the literature reported models that included the similar and less diverse molecules. Our models can be used in identifying the potentially carcinogenic chemicals for a wide variety of chemical classes.

2 citations

Proceedings ArticleDOI
28 Apr 2020
TL;DR: Methods to suppress clutter and to extract features of weak radar targets, such as pedestrians, are discussed and algorithms to detect pedestrians in a high clutter environment can be investigated with the help of MATLAB® simulation tools.
Abstract: Embedded micro radars that include integrated hardware, application software, and a real-time operating system as part of an enclosed system provide convenient platforms for outdoor and indoor environment monitoring. This paper discusses methods to suppress clutter and to extract features of weak radar targets, such as pedestrians. With the help of MATLAB® simulation tools, algorithms to detect pedestrians in a high clutter environment can be investigated. A hybrid neural network is developed on micro-Doppler signatures generated synthetically. It outperforms the traditional convolutional neural network in terms of accuracy and the amount of training data required. We then use an embedded micro radar for field trials to test the performance of the hybrid neural network on real data sets.

2 citations


Network Information
Related Topics (5)
Artificial neural network
207K papers, 4.5M citations
89% related
Feature extraction
111.8K papers, 2.1M citations
88% related
Fuzzy logic
151.2K papers, 2.3M citations
85% related
Convolutional neural network
74.7K papers, 2M citations
84% related
Deep learning
79.8K papers, 2.1M citations
83% related
Performance
Metrics
No. of papers in the topic in previous years
YearPapers
20233
20228
2021128
2020119
2019104
201863