A simple feedforward convolutional conceptor neural network for classification

Qian, Guangwu; Zhang, Lei

doi:10.1016/j.asoc.2017.08.016

Cited by 33 publications

(20 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The classification results of these 20 methods are collected from the corresponding papers. These methods are SVM+RBF [34], SVM+Poly [34], SAE-3 [33], DAE-b-3 [33], CAE-2 [33], SPAE [48], RBM-3 [33], ScatNet-2 [31,32], RandNet-2 [32], PCANet-2 (softmax) [32], LDANet-2 [32], NNet [34], SAA-3 [34], DBN-3 [34], FCCNN [29], FCCNN (with BT) [29], SPCN [30], EvoCNN [49], EGP [37], and IEGP [38]. Most of these methods are NN-based methods, which have been introduced in Section II-B2.…”

Section: B Benchmark Methodsmentioning

confidence: 99%

“…Among these methods, CNN is the most commonly used method for image classification. Qian and Zhang [29] developed a feedforward convolutional conceptor neural network (FCCNN) by integrating components of CNN, principal component analysis (PCA), binary thresholding (BT) and nontemporal conceptor classifiers. The performance of FCCNN has been examined on the MNIST variant datasets.…”

Section: B Related Work On Image Classificationmentioning

confidence: 99%

See 1 more Smart Citation

Genetic Programming With Image-Related Operators and a Flexible Program Structure for Feature Learning in Image Classification

Xue

Zhang

2021

IEEE Trans. Evol. Computat.

View full text Add to dashboard Cite

Feature extraction is essential for solving image classification by transforming low-level pixel values into highlevel features. However, extracting effective features from images is challenging due to high variations across images in scale, rotation, illumination, and background. Existing methods often have a fixed model complexity and require domain expertise. Genetic programming with a flexible representation can find the best solution without the use of domain knowledge. This paper proposes a new genetic programming-based approach to automatically learning informative features for different image classification tasks. In the new approach, a number of imagerelated operators, including filters, pooling operators and feature extraction methods, are employed as functions. A flexible program structure is developed to integrate different functions and terminals into a single tree/solution. The new approach can evolve solutions of variable depths to extract various numbers and types of features from the images. The new approach is examined on 12 different image classification tasks of varying difficulty and compared with a large number of effective algorithms. The results show that the new approach achieves better classification performance than most benchmark methods. The analysis of the evolved programs/solutions and the visualisation of the learned features provide deep insights on the proposed approach.

show abstract

Section: B Benchmark Methodsmentioning

confidence: 99%

Section: B Related Work On Image Classificationmentioning

confidence: 99%

Genetic Programming With Image-Related Operators and a Flexible Program Structure for Feature Learning in Image Classification

Xue

Zhang

2021

IEEE Trans. Evol. Computat.

View full text Add to dashboard Cite

show abstract

“…We have collected these results from corresponding references on data sets 7-13. Therefore, there are 19 comparison methods on data sets 7-13, i.e., SVM+RBF [50], SVM+Poly [50], SAE-3 [35], DAE-b-3 [35], CAE-2 [35], SPAE [51], RBM-3 [35], ScatNet-2 [32,33], RandNet-2 [33], PCANet-2 (softmax) [33], LDANet-2 [33], NNet [50], SAA-3 [50], DBN-3 [50], FCCNN [34], FCCNN (with BT) [34], SPCN [31], EGP [25], and EvoCNN [52]. Note that most of these methods are neural network-based methods and parameter tuning was conducted in some methods to obtain a good classification performance.…”

Section: B Benchmark Methodsmentioning

confidence: 99%

“…Several variants of CNNs have been developed, such as scattering convolution networks (ScatNet) [32] and principal component analysis network (PCANet) [33], where the convolution filters are generated using different ways. In [34], a feedforward convolutional conceptor neural network (FCCNN) was developed for image classification. The contractive auto-encoder (CAE) was employed by Rifai et al [35] for digits recognition.…”

Section: Image Feature Extraction and Learningmentioning

confidence: 99%

Genetic Programming With a New Representation to Automatically Learn Features and Evolve Ensembles for Image Classification

Xue

Zhang

2021

IEEE Trans. Cybern.

View full text Add to dashboard Cite

Image classification is a popular task in machine learning and computer vision, but it is very challenging due to high variations crossing images. Using ensemble methods for solving image classification can achieve higher classification performance than using a single classification algorithm. However, to obtain a good ensemble, the component (base) classifiers in an ensemble should be accurate and diverse. To solve image classification effectively, feature extraction is necessary to transform raw pixels into high-level informative features. However, this process often requires domain knowledge. This paper proposes an evolutionary approach based on genetic programming to automatically and simultaneously learn informative features and evolve effective ensembles for image classification. The new approach takes raw images as inputs and returns predictions of class labels based on the evolved classifiers. To achieve this, a new individual representation, a new function set and a new terminal set are developed to allow the new approach to effectively find the best solution. More importantly, the solutions of the new approach can extract informative features from raw images and can automatically address the diversity issue of the ensembles. In addition, the new approach can automatically select and optimise the parameters for the classification algorithms in the ensemble. The performance of the new approach is examined on 13 different image classification data sets of varying difficulty and compared with a large number of effective methods. The results show that the new approach achieves better classification accuracy on most data sets than the competitive methods. Further analysis demonstrates that the new approach can evolve solutions with high accuracy and diversity.

show abstract

“…Conceptor networks have the significant advantage that they are capable of dynamically loading new classes of data patterns, a difficult problem for most other conventional supervised training-based classifiers which can fail to recognize new classes of the input pattern. Because of these advantages, Conceptor networks have been applied to fields such as image classification [23], [24], natural language processing [25], time series classification [26].…”

Section: Introductionmentioning

confidence: 99%

Phase Space Reconstruction-Based Conceptor Network for Time Series Prediction

2019

View full text Add to dashboard Cite

The Conceptor network is a newly proposed reservoir computing (RC) model, which outperforms traditional classifiers, which can fail to model new classes of data for a supervised learning task. However, the reservoir structure design for the Conceptor is single, involving just a traditional random network, which has strong coupling between nodes and limits computing ability. This study focused on the reservoir topology design problem, and we propose a complex network Conceptor-based phase space reconstruction of time series. Several dynamical systems were chosen to build complex networks using a phase space reconstruction algorithm. The experiment results obtained using a mix of two irrational-period sines showed that the proposed phase space reconstruction reservoir topologies with the appropriate values of threshold provide Conceptors with extra reconstruction precision. Among them, the phase space reconstruction reservoir-based Lorenz system shows the best performance. Further experiments also identified the appropriate values of threshold of the phase space reconstruction method required to obtain optimal performance. The precision showed a non-linear decline with increase in memory load, and the proposed Lorenz phase space reconstruction reservoir maintained its advantages under different memory loads. INDEX TERMS Conceptor, reservoir computing, phase space reconstruction, time series prediction.

show abstract

A simple feedforward convolutional conceptor neural network for classification

Cited by 33 publications

References 19 publications

Genetic Programming With Image-Related Operators and a Flexible Program Structure for Feature Learning in Image Classification

Genetic Programming With Image-Related Operators and a Flexible Program Structure for Feature Learning in Image Classification

Genetic Programming With a New Representation to Automatically Learn Features and Evolve Ensembles for Image Classification

Phase Space Reconstruction-Based Conceptor Network for Time Series Prediction

Contact Info

Product

Resources

About