Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033357
A. R. Rao, G. Cecchi
We model neural dynamical behavior during object perception using the principle of sparse coding in multilayer oscillatory networks. The network model consists of units with amplitude and phase variables, and allows the propagation of higher-level information to lower levels via feedback connections. We show that this model can replicate findings in the neuroscience literature, where measurements have shown that neurons in lower level visual areas respond in a delayed fashion to missing contours of whole objects. We contrast the behavior of feedback connections with that of lateral connections by selectively disabling these in our model to examine their contributions to object perception. This paper successfully extends the previously reported capabilities of oscillatory networks by applying them to model perceptual tasks.
{"title":"The effects of feedback and lateral connections on perceptual processing: A study using oscillatory networks","authors":"A. R. Rao, G. Cecchi","doi":"10.1109/IJCNN.2011.6033357","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033357","url":null,"abstract":"We model neural dynamical behavior during object perception using the principle of sparse coding in multilayer oscillatory networks. The network model consists of units with amplitude and phase variables, and allows the propagation of higher-level information to lower levels via feedback connections. We show that this model can replicate findings in the neuroscience literature, where measurements have shown that neurons in lower level visual areas respond in a delayed fashion to missing contours of whole objects. We contrast the behavior of feedback connections with that of lateral connections by selectively disabling these in our model to examine their contributions to object perception. This paper successfully extends the previously reported capabilities of oscillatory networks by applying them to model perceptual tasks.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124866918","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033653
M. El-Melegy
This paper addresses the problem of fitting a functional model to data corrupted with outliers using a multilayered feed-forward neural network (MFNN). Almost all previous efforts to solve this problem have focused on using a training algorithm that minimizes an M-estimator based error criterion. However the robustness gained from M-estimators is still low. Using a training algorithm based on the RANdom SAmple Consensus (RANSAC) framework improves significantly the robustness of the algorithm. However the algorithm typically requires prolonged period of time before a final solution is reached. In this paper, we propose a new strategy to improve the time performance of the RANSAC algorithm for training MFNNs. A statistical pre-test based on Wald's sequential probability ratio test (SPRT) is performed on each randomly generated sample to decide whether it deserves to be used for model estimation. The proposed algorithm is evaluated on synthetic data, contaminated with varying degrees of outliers, and have demonstrated faster performance compared to the original RANSAC algorithm with no significant sacrifice of the robustness.
{"title":"RANSAC algorithm with sequential probability ratio test for robust training of feed-forward neural networks","authors":"M. El-Melegy","doi":"10.1109/IJCNN.2011.6033653","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033653","url":null,"abstract":"This paper addresses the problem of fitting a functional model to data corrupted with outliers using a multilayered feed-forward neural network (MFNN). Almost all previous efforts to solve this problem have focused on using a training algorithm that minimizes an M-estimator based error criterion. However the robustness gained from M-estimators is still low. Using a training algorithm based on the RANdom SAmple Consensus (RANSAC) framework improves significantly the robustness of the algorithm. However the algorithm typically requires prolonged period of time before a final solution is reached. In this paper, we propose a new strategy to improve the time performance of the RANSAC algorithm for training MFNNs. A statistical pre-test based on Wald's sequential probability ratio test (SPRT) is performed on each randomly generated sample to decide whether it deserves to be used for model estimation. The proposed algorithm is evaluated on synthetic data, contaminated with varying degrees of outliers, and have demonstrated faster performance compared to the original RANSAC algorithm with no significant sacrifice of the robustness.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125009023","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033592
Shozo Shimomura, H. Matsushita, Y. Nishio
In our previous study, we have proposed an Ant Colony Optimization with Intelligent and Dull Ants (IDACO) which contains two kinds of ants. We have applied IDACO to various Traveling Salesman Problems (TSPs) and confirmed its effectiveness. This study proposes an Ant Colony Optimization Changing the Rate of Dull Ants (IDACO-CR) and its Application to Quadratic Assignment Problems (QAPs). In addition to the existence of the dull ants which cannot trail the pheromone, the rate of dull ants in IDACO-CR is changed flexibly and automatically in the simulation, depending on the problem. We investigate the behavior of IDACO-CR in detail and the effect of changing the rate of dull ants. Simulation results show that IDACO-CR gets out from the local optima by changing the rate of dull ants, and we confirm that IDACO-CR obtains the effective results in solving complex optimization problems.
{"title":"Ant Colony Optimization Changing the Rate of Dull Ants and its application to QAP","authors":"Shozo Shimomura, H. Matsushita, Y. Nishio","doi":"10.1109/IJCNN.2011.6033592","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033592","url":null,"abstract":"In our previous study, we have proposed an Ant Colony Optimization with Intelligent and Dull Ants (IDACO) which contains two kinds of ants. We have applied IDACO to various Traveling Salesman Problems (TSPs) and confirmed its effectiveness. This study proposes an Ant Colony Optimization Changing the Rate of Dull Ants (IDACO-CR) and its Application to Quadratic Assignment Problems (QAPs). In addition to the existence of the dull ants which cannot trail the pheromone, the rate of dull ants in IDACO-CR is changed flexibly and automatically in the simulation, depending on the problem. We investigate the behavior of IDACO-CR in detail and the effect of changing the rate of dull ants. Simulation results show that IDACO-CR gets out from the local optima by changing the rate of dull ants, and we confirm that IDACO-CR obtains the effective results in solving complex optimization problems.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125064104","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Neural clustering algorithms show high performance in the general context of the analysis of homogeneous textual dataset. This is especially true for the recent adaptive versions of these algorithms, like the incremental growing neural gas algorithm (IGNG) and the labeling maximization based incremental growing neural gas algorithm (IGNG-F). In this paper we highlight that there is a drastic decrease of performance of these algorithms, as well as the one of more classical algorithms, when a heterogeneous textual dataset is considered as an input. Specific quality measures and cluster labeling techniques that are independent of the clustering method are used for the precise performance evaluation. We provide new variations to incremental growing neural gas algorithm exploiting in an incremental way knowledge from clusters about their current labeling along with cluster distance measure data. This solution leads to significant gain in performance for all types of datasets, especially for the clustering of complex heterogeneous textual data.
{"title":"Variations to incremental growing neural gas algorithm based on label maximization","authors":"Jean-Charles Lamirel, Raghvendra Mall, Pascal Cuxac, Ghada Safi","doi":"10.1109/IJCNN.2011.6033326","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033326","url":null,"abstract":"Neural clustering algorithms show high performance in the general context of the analysis of homogeneous textual dataset. This is especially true for the recent adaptive versions of these algorithms, like the incremental growing neural gas algorithm (IGNG) and the labeling maximization based incremental growing neural gas algorithm (IGNG-F). In this paper we highlight that there is a drastic decrease of performance of these algorithms, as well as the one of more classical algorithms, when a heterogeneous textual dataset is considered as an input. Specific quality measures and cluster labeling techniques that are independent of the clustering method are used for the precise performance evaluation. We provide new variations to incremental growing neural gas algorithm exploiting in an incremental way knowledge from clusters about their current labeling along with cluster distance measure data. This solution leads to significant gain in performance for all types of datasets, especially for the clustering of complex heterogeneous textual data.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"147 4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125867977","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033580
Hakan Yasarer, Y. Najjar
Corrosion of reinforcing steel due to chloride penetration is one of the most common causes of deterioration in concrete pavement structures. On an annual basis, millions of dollars are spent on corrosion-related repairs. High incidence rates and repair costs have stimulated widespread research interests in order to properly assess the durability problem of concrete pavements. Chloride penetration of concrete pavement structures is determined through the Rapid Chloride Permeability test (RCPT), which typically measures the number of coulombs passing through a concrete sample over a period of six hours at a concrete age of 7, 28, and 56 days. In a composite material, such as concrete, the parameters of the mixture design and interaction between them determine the behavior of the material. Previous studies have shown that Artificial Neural Network (ANN) based material modeling approach has been successfully used to capture complex interactions among input and output variables. In this study, back-propagation ANN, and Regression-based permeability response prediction models were developed to assess the permeability potential of various concrete mixes using data obtained from actual Rapid Chloride Permeability tests. The back-propagation ANN learning technique proved to be an efficient method to produce relatively accurate permeability response prediction models. Comparison of the prediction accuracy of the developed ANN models and the regression model proved that the developed ANN model outperformed the regression-based model. The developed ANN models have high predictive capability to properly assess the chloride permeability of concrete mixes based on various mix-design parameters. These models can reliably be used for permeability prediction tasks in order to reduce or eliminate the duration of the testing as well as the sample preparation periods required for proper RCP testing.
{"title":"Development of a mix-design based Rapid Chloride Permeability assessment model using neuronets","authors":"Hakan Yasarer, Y. Najjar","doi":"10.1109/IJCNN.2011.6033580","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033580","url":null,"abstract":"Corrosion of reinforcing steel due to chloride penetration is one of the most common causes of deterioration in concrete pavement structures. On an annual basis, millions of dollars are spent on corrosion-related repairs. High incidence rates and repair costs have stimulated widespread research interests in order to properly assess the durability problem of concrete pavements. Chloride penetration of concrete pavement structures is determined through the Rapid Chloride Permeability test (RCPT), which typically measures the number of coulombs passing through a concrete sample over a period of six hours at a concrete age of 7, 28, and 56 days. In a composite material, such as concrete, the parameters of the mixture design and interaction between them determine the behavior of the material. Previous studies have shown that Artificial Neural Network (ANN) based material modeling approach has been successfully used to capture complex interactions among input and output variables. In this study, back-propagation ANN, and Regression-based permeability response prediction models were developed to assess the permeability potential of various concrete mixes using data obtained from actual Rapid Chloride Permeability tests. The back-propagation ANN learning technique proved to be an efficient method to produce relatively accurate permeability response prediction models. Comparison of the prediction accuracy of the developed ANN models and the regression model proved that the developed ANN model outperformed the regression-based model. The developed ANN models have high predictive capability to properly assess the chloride permeability of concrete mixes based on various mix-design parameters. These models can reliably be used for permeability prediction tasks in order to reduce or eliminate the duration of the testing as well as the sample preparation periods required for proper RCP testing.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126068594","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033554
Yi-Ren Yeh, Y. Chung, Ting-Chu Lin, Y. Wang
We propose a novel multiple kernel learning (MKL) algorithm with a group lasso regularizer, called group lasso regularized MKL (GL-MKL), for heterogeneous feature selection. We extend the existing MKL algorithm and impose a mixed ℓ1 and ℓ2 norm constraint (known as group lasso) as the regularizer. Our GL-MKL determines the optimal base kernels, including the associated weights and kernel parameters, and results in a compact set of features for comparable or improved recognition performance. The use of our GL-MKL avoids the problem of choosing the proper technique to normalize the feature attributes collected from heterogeneous domains (and thus with different properties and distribution ranges). Our approach does not need to exhaustively search for the entire feature space when performing feature selection like prior sequential-based feature selection methods did, and we do not require any prior knowledge on the optimal size of the feature subset either. Comparisons with existing MKL or sequential-based feature selection methods on a variety of datasets confirm the effectiveness of our method in selecting a compact feature subset for comparable or improved classification performance.
{"title":"Group lasso regularized multiple kernel learning for heterogeneous feature selection","authors":"Yi-Ren Yeh, Y. Chung, Ting-Chu Lin, Y. Wang","doi":"10.1109/IJCNN.2011.6033554","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033554","url":null,"abstract":"We propose a novel multiple kernel learning (MKL) algorithm with a group lasso regularizer, called group lasso regularized MKL (GL-MKL), for heterogeneous feature selection. We extend the existing MKL algorithm and impose a mixed ℓ1 and ℓ2 norm constraint (known as group lasso) as the regularizer. Our GL-MKL determines the optimal base kernels, including the associated weights and kernel parameters, and results in a compact set of features for comparable or improved recognition performance. The use of our GL-MKL avoids the problem of choosing the proper technique to normalize the feature attributes collected from heterogeneous domains (and thus with different properties and distribution ranges). Our approach does not need to exhaustively search for the entire feature space when performing feature selection like prior sequential-based feature selection methods did, and we do not require any prior knowledge on the optimal size of the feature subset either. Comparisons with existing MKL or sequential-based feature selection methods on a variety of datasets confirm the effectiveness of our method in selecting a compact feature subset for comparable or improved classification performance.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123261262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033387
Alberione Braz da Silva, J. Rosa
Artificial neural network (ANN) community is engaged in biological plausibility issues these days. Different views about this subject can lead to disagreements of classification criteria among ANN researchers. In order to contribute to this debate, two of these views are highlighted here: one is related directly to the cerebral cortex biological structure, and the other focuses the neural features and the signaling between neurons. The model proposed in this paper considers that a biologically more plausible ANN has the purpose to create a more faithful model concerning the biological structure, properties, and functionalities, including learning processes, of the cerebral cortex, not disregarding its computational efficiency. The choice of the models upon which the proposed description is based takes into account two main criteria: the fact they are considered biologically more realistic and the fact they deal with intra and inter-neuron signaling in electrical and chemical synapses. Also, the duration of action potentials is taken into account. In addition to the characteristics for encoding information regarding biological plausibility present in current spiking neuron models, a distinguishable feature is emphasized here: a combination of Hebbian learning and error-driven learning.
{"title":"Advances on criteria for biological plausibility in artificial neural networks: Think of learning processes","authors":"Alberione Braz da Silva, J. Rosa","doi":"10.1109/IJCNN.2011.6033387","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033387","url":null,"abstract":"Artificial neural network (ANN) community is engaged in biological plausibility issues these days. Different views about this subject can lead to disagreements of classification criteria among ANN researchers. In order to contribute to this debate, two of these views are highlighted here: one is related directly to the cerebral cortex biological structure, and the other focuses the neural features and the signaling between neurons. The model proposed in this paper considers that a biologically more plausible ANN has the purpose to create a more faithful model concerning the biological structure, properties, and functionalities, including learning processes, of the cerebral cortex, not disregarding its computational efficiency. The choice of the models upon which the proposed description is based takes into account two main criteria: the fact they are considered biologically more realistic and the fact they deal with intra and inter-neuron signaling in electrical and chemical synapses. Also, the duration of action potentials is taken into account. In addition to the characteristics for encoding information regarding biological plausibility present in current spiking neuron models, a distinguishable feature is emphasized here: a combination of Hebbian learning and error-driven learning.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123543839","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033392
Nathan Rose
Radial Basis Networks are a reliable and efficient tool for performing classification tasks. In networks that include a Gaussian output transform within the Pattern Layer neurons, the method of setting the σ-width of the Gaussian curve is critical to obtaining accurate classification. Many existing methods perform poorly in regions of the problem space between examples of differing classes, or when there is overlap between classes in the data set. A method is proposed to produce unique σ values for each weight of every neuron, resulting in each neuron having its own Gaussian ‘coverage’ area within problem space. This method achieves better results than the alternatives on data sets with a significant amount of overlap and when the data is unscaled.
{"title":"Selective adjustment of rotationally-asymmetric neuron σ-widths","authors":"Nathan Rose","doi":"10.1109/IJCNN.2011.6033392","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033392","url":null,"abstract":"Radial Basis Networks are a reliable and efficient tool for performing classification tasks. In networks that include a Gaussian output transform within the Pattern Layer neurons, the method of setting the σ-width of the Gaussian curve is critical to obtaining accurate classification. Many existing methods perform poorly in regions of the problem space between examples of differing classes, or when there is overlap between classes in the data set. A method is proposed to produce unique σ values for each weight of every neuron, resulting in each neuron having its own Gaussian ‘coverage’ area within problem space. This method achieves better results than the alternatives on data sets with a significant amount of overlap and when the data is unscaled.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123672715","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033285
V. Cherkassky, S. Chowdhury, Volker Landenberger, Saurabh Tewari, P. Bursch
Currently many commercial buildings are not continuously monitored for energy consumption, especially small buildings which constitute 90% of all such buildings. However, readily available data from the electric meters can be used for monitoring and analyzing energy consumption. Efficient utilization of available historical data (from these meters) can potentially improve energy efficiency, help to identify common energy wasting problems, and, in the future, enable various Smart Grid programs, such as demand response, real-time pricing etc. This paper describes application of computational intelligence techniques for prediction of electric power consumption. The proposed approach combines regression and clustering methods, in order to improve the prediction accuracy of power consumption, as a function of time (of the day) and temperature, using real-life data from several commercial and government buildings. Empirical comparisons show that the proposed approach provides an improvement over the currently used bin-based method for modeling power consumption.
{"title":"Prediction of electric power consumption for commercial buildings","authors":"V. Cherkassky, S. Chowdhury, Volker Landenberger, Saurabh Tewari, P. Bursch","doi":"10.1109/IJCNN.2011.6033285","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033285","url":null,"abstract":"Currently many commercial buildings are not continuously monitored for energy consumption, especially small buildings which constitute 90% of all such buildings. However, readily available data from the electric meters can be used for monitoring and analyzing energy consumption. Efficient utilization of available historical data (from these meters) can potentially improve energy efficiency, help to identify common energy wasting problems, and, in the future, enable various Smart Grid programs, such as demand response, real-time pricing etc. This paper describes application of computational intelligence techniques for prediction of electric power consumption. The proposed approach combines regression and clustering methods, in order to improve the prediction accuracy of power consumption, as a function of time (of the day) and temperature, using real-life data from several commercial and government buildings. Empirical comparisons show that the proposed approach provides an improvement over the currently used bin-based method for modeling power consumption.","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123674517","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2011-10-03DOI: 10.1109/IJCNN.2011.6033589
P. Sermanet, Yann LeCun
We apply Convolutional Networks (ConvNets) to the task of traffic sign classification as part of the GTSRB competition. ConvNets are biologically-inspired multi-stage architectures that automatically learn hierarchies of invariant features. While many popular vision approaches use hand-crafted features such as HOG or SIFT, ConvNets learn features at every level from data that are tuned to the task at hand. The traditional ConvNet architecture was modified by feeding 1st stage features in addition to 2nd stage features to the classifier. The system yielded the 2nd-best accuracy of 98.97% during phase I of the competition (the best entry obtained 98.98%), above the human performance of 98.81%, using 32×32 color input images. Experiments conducted after phase 1 produced a new record of 99.17% by increasing the network capacity, and by using greyscale images instead of color. Interestingly, random features still yielded competitive results (97.33%).
{"title":"Traffic sign recognition with multi-scale Convolutional Networks","authors":"P. Sermanet, Yann LeCun","doi":"10.1109/IJCNN.2011.6033589","DOIUrl":"https://doi.org/10.1109/IJCNN.2011.6033589","url":null,"abstract":"We apply Convolutional Networks (ConvNets) to the task of traffic sign classification as part of the GTSRB competition. ConvNets are biologically-inspired multi-stage architectures that automatically learn hierarchies of invariant features. While many popular vision approaches use hand-crafted features such as HOG or SIFT, ConvNets learn features at every level from data that are tuned to the task at hand. The traditional ConvNet architecture was modified by feeding 1st stage features in addition to 2nd stage features to the classifier. The system yielded the 2nd-best accuracy of 98.97% during phase I of the competition (the best entry obtained 98.98%), above the human performance of 98.81%, using 32×32 color input images. Experiments conducted after phase 1 produced a new record of 99.17% by increasing the network capacity, and by using greyscale images instead of color. Interestingly, random features still yielded competitive results (97.33%).","PeriodicalId":415833,"journal":{"name":"The 2011 International Joint Conference on Neural Networks","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2011-10-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114958133","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：