FlexConv: Continuous Kernel Convolutions with Differentiable Kernel Sizes
arXiv:2110.08059
Abstract
When designing Convolutional Neural Networks (CNNs), one must select the size\break of the convolutional kernels before training. Recent works show CNNs benefit from different kernel sizes at different layers, but exploring all possible combinations is unfeasible in practice. A more efficient approach is to learn the kernel size during training. However, existing works that learn the kernel size have a limited bandwidth. These approaches scale kernels by dilation, and thus the detail they can describe is limited. In this work, we propose FlexConv, a novel convolutional operation with which high bandwidth convolutional kernels of learnable kernel size can be learned at a fixed parameter cost. FlexNets model long-term dependencies without the use of pooling, achieve state-of-the-art performance on several sequential datasets, outperform recent works with learned kernel sizes, and are competitive with much deeper ResNets on image benchmark datasets. Additionally, FlexNets can be deployed at higher resolutions than those seen during training. To avoid aliasing, we propose a novel kernel parameterization with which the frequency of the kernels can be analytically controlled. Our novel kernel parameterization shows higher descriptive power and faster convergence speed than existing parameterizations. This leads to important improvements in classification accuracy.
First two authors contributed equally to this work
References in corpus (14)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- Neural Architecture Search with Reinforcement Learning
- Alias-Free Generative Adversarial Networks
- A Simple Way to Initialize Recurrent Networks of Rectified Linear Units
- Making Convolutional Networks Shift-Invariant Again
- A Downsampled Variant of ImageNet as an Alternative to the CIFAR datasets
- Efficient-CapsNet: Capsule Network with Self-Attention Routing
- AntisymmetricRNN: A Dynamical System View on Recurrent Neural Networks
- Blurring the Line Between Structure and Learning to Optimize and Adapt Receptive Fields
- CKConv: Continuous Kernel Convolution For Sequential Data
- Parallelizing Legendre Memory Unit Training
- UnICORNN: A recurrent model for learning very long time dependencies
- Deep Continuous Networks