This article proposes a novel method for one-class classification based on a divide-and-conquer strategy to improve the one-class support vector machine (SVM). The idea is to build a piecewise linear separation bounda...
详细信息
This article proposes a novel method for one-class classification based on a divide-and-conquer strategy to improve the one-class support vector machine (SVM). The idea is to build a piecewise linear separation boundary in the feature space to separate the data points from the origin, which is expected to have a more compact region in the input space. For the purpose, the input space of the dataset is first divided into a group of partitions by using a partitioning mechanism of top s% winner-take-all autoencoder. A gated linear network is designed to implement a group of linear classifiers for each partition, in which the gate signals are generated from the autoencoder. By applying a one-class SVM (OCSVM) formulation to optimize the parameter set of the gated linear network, the one-class classifier is implemented in an exactly same way as a standard OCSVM with a quasi-linear kernel composed using a base kernel with the gate signals. The proposed one-class classification method is applied to different real-world datasets, and simulation results show that it shows a better performance than a traditional OCSVM. (c) 2018 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.
In this paper, we propose a new mid-level visual elements discovery method and apply it to the fine-grained classification. We present the duality between image patches and features extracted by the convolutional winn...
详细信息
In this paper, we propose a new mid-level visual elements discovery method and apply it to the fine-grained classification. We present the duality between image patches and features extracted by the convolutional winner-take-all autoencoder (CONV-WTA-AE). The sparsity constraints used by CONV-WTA-AE make a group of objects sharing the same feature components. Hence, the image patches could be clustered by their sharing feature components and the feature components can be clustered by their co-occurrence in the image patches. We propose formulating the mid-level visual elements mining as a bipartite graph partitioning problem. The spectral partitioning algorithm is employed to co-cluster image patches and feature components. The CONV-WTA-AE is an unsupervised feature learning method. Hence, it avoids using expensive annotations. Our experiments demonstrate that the spectral partitioning method is very efficient but only the confident instances in a cluster are well discriminated. The similarity metric used by this algorithm is not accurate enough. Hence, we propose training a group of linear support vector machine (SVM) to refine the clustering results. These SVMs will be trained on the initial confident instances and provide a better discriminative similarity. Then we can re-assign instances to each clusters. To avoid overfitting, this process is iterated on many data subsets. We conduct a series of experiments on the MNIST dataset to verify our algorithm. The experimental results show that our method can discover meaningful image patch clusters. In the fine-grained classification task, visual elements are input into an ensemble of convolutional neural networks. The experiments on the CompCars dataset illustrate that our method can achieve the state-of-the-art performance.
Air pollution has threaten people's health. It is urgent for the government to strengthen and enhance air pollution monitoring capacity. In this paper, we propose an air quality prediction model to infer air pollu...
详细信息
ISBN:
(纸本)9781538640883
Air pollution has threaten people's health. It is urgent for the government to strengthen and enhance air pollution monitoring capacity. In this paper, we propose an air quality prediction model to infer air pollutant concentrations, such as CO, NOx, NO2. The idea is to design a sophisticate piecewise linear model by using a gated linear network. A top k% winner-take-all autoencoder is first built to generate a set of binary sequences as the gate control signals, so as to perform the input space partitioning. The piecewise linear model is then identified in an exact same way as a standard support vector regression (SVR) with a quasi-linear kernel composed by using the gate control signals. Results of our experiments shows that our proposed SVR prediction model outperforms other state-of-the-art methods.
暂无评论