Abstract:In recent years, Convolutional Neural Networks (CNNs) have enabled ubiquitous image processing applications. As such, CNNs require fast runtime (forward propagation) to process high-resolution visual streams in real time. This is still a challenging task even with state-of-the-art graphics and tensor processing units. The bottleneck in computational efficiency primarily occurs in the convolutional layers. Performing operations in the Fourier domain is a promising way to accelerate forward propagation since it transforms convolutions into elementwise multiplications, which are considerably faster to compute for large kernels. Furthermore, such computation could be implemented using an optical 4f system with orders of magnitude faster operation. However, a major challenge in using this spectral approach, as well as in an optical implementation of CNNs, is the inclusion of a nonlinearity between each convolutional layer, without which CNN performance drops dramatically. Here, we propose a Spectral CNN Linear Counterpart (SCLC) network architecture and develop a Knowledge Distillation (KD) approach to circumvent the need for a nonlinearity and successfully train such networks. While the KD approach is known in machine learning as an effective process for network pruning, we adapt the approach to transfer the knowledge from a nonlinear network (teacher) to a linear counterpart (student). We show that the KD approach can achieve performance that easily surpasses the standard linear version of a CNN and could approach the performance of the nonlinear network. Our simulations show that the possibility of increasing the resolution of the input image allows our proposed 4f optical linear network to perform more efficiently than a nonlinear network with the same accuracy on two fundamental image processing tasks: (i) object classification and (ii) semantic segmentation.

Non-linear Convolution Filters for CNN-based Learning

Advances in Convolutional Neural Networks

Nonlinear CNN: Improving CNNs with Quadratic Convolutions

Convolution goes higher-order: a biologically inspired mechanism empowers image classification

Convolutional Neural Network with Nonlinear Competitive Units.

On the Learning Dynamics of Two-layer Nonlinear Convolutional Neural Networks.

Elliptical Convolution Kernel: More Real Visual Field

Kervolutional Neural Networks

Knowledge Distillation Circumvents Nonlinearity for Optical Convolutional Neural Networks

A Tour of Convolutional Networks Guided by Linear Interpreters

Volume Upscaling with Convolutional Neural Networks

Convolutional networks and applications in vision

Efficient Higher-order Convolution for Small Kernels in Deep Learning

Convolutional Neural Networks Exploiting Attributes of Biological Neurons

Convolutional neural network models of neuronal responses in macaque V1 reveal limited non-linear processing

Convolutional neural networks for vision neuroscience: significance, developments, and outstanding issues

Learning Filter Scale and Orientation In CNNs

Convolutional neural network models applied to neuronal responses in macaque V1 reveal limited nonlinear processing

Optical Convolutional Neural Networks -- Combining Silicon Photonics and Fourier Optics for Computer Vision

Brain inspired Robust Vision using Convolutional Neural Networks with Feedback

Efficient and Accurate Approximations of Nonlinear Convolutional Networks