Learning Deep and Wide: A Spectral Method for Learning Deep Networks

By Ling Shao, Xuelong Li. 2014

144 citations recorded by the source corpus.

Building intelligent systems that are capable of extracting high-level representations from high-dimensional sensory data lies at the core of solving many computer vision-related tasks. We propose the multispectral neural networks (MSNN) to learn features from multicolumn deep neural networks and embed the penultimate hierarchical discriminative manifolds into a compact representation. The low-dimensional embedding explores the complementary property of different views wherein the distribution of each view is sufficiently smooth and hence achieves robustness, given few labeled training data. Our experiments show that spectrally embedding several deep neural networks can explore the optimum output from the multicolumn networks and consistently decrease the error rate compared with a single deep network.

Subfields: Computer Vision and Pattern Recognition, Media Technology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Dacheng Tao
2. Liangpei Zhang
3. Bo Du
4. Xuelong Li
5. Lefei Zhang
6. Qi Wang
7. Yuan Yuan
8. Xinbo Gao

In Media Technology

1. Liangpei Zhang
2. Bo Du
3. Xuelong Li
4. Dacheng Tao
5. Lefei Zhang
6. Qi Wang
7. Yuan Yuan
8. Jianzhe Lin

Citations in the corpus

• Stacked Convolutional Denoising Auto-Encoders for Feature Representation (2017, methodology): “network (CDBN), which replaces the RBM in each layer with CRBM, and uses convolution structure to combine the layers to construct hierarchical models [29], [30].”
• Blind Image Quality Assessment via Deep Learning (2015, background): “Deep learning models have thrived over the years [30], not only in the field of computer vision, but also in many others, such as audio [31], natural language processing [32], and so on [21], [33]–[35].”
• Deep Neural Network for Structural Prediction and Lane Detection in Traffic Scene (2017, background): “3 and 5, respectively. has led to breakthrough in effective learning schemes [9]–[12].”
• Weakly supervised target detection in remote sensing images based on transferred deep features and negative bootstrapping (2015, unknown): “To address the shortcomings of handcrafted features, some high-level image representations (Shao et al. 2014b; Zhang et al. 2014) and domain-adaptive learning methods (Shao et al. 2014a; Zhu and Shao 2014) have been proposed.”
• On building ensembles of stacked denoising auto-encoding classifiers and their further improvement (2018, background): “The authors of [58] use spectral diversity [59] to train some DNNs whose outputs are aggregated, while partial training of learners with distorted data sets in the diversifi-cation applied in [60] .”
• Salient Band Selection for Hyperspectral Image Classification via Manifold Ranking (2016, methodology): “In this section, we will talk about the state of the art classifier—deep neural networks [50], [51] for HSI classification combined with our band selection process.”