Compressed Learning: A Deep Neural Network Approach
This work addresses the challenge of efficient inference from limited measurements for signal processing applications, representing an incremental improvement over existing compressed learning techniques.
The paper tackles the problem of compressed learning for image classification by proposing an end-to-end deep neural network that jointly optimizes linear sensing and non-linear inference, achieving a classification error of 6.46% on MNIST at a 1% sensing rate compared to 41.06% with state-of-the-art methods.
Compressed Learning (CL) is a joint signal processing and machine learning framework for inference from a signal, using a small number of measurements obtained by linear projections of the signal. In this paper we present an end-to-end deep learning approach for CL, in which a network composed of fully-connected layers followed by convolutional layers perform the linear sensing and non-linear inference stages. During the training phase, the sensing matrix and the non-linear inference operator are jointly optimized, and the proposed approach outperforms state-of-the-art for the task of image classification. For example, at a sensing rate of 1% (only 8 measurements of 28 X 28 pixels images), the classification error for the MNIST handwritten digits dataset is 6.46% compared to 41.06% with state-of-the-art.