Kernel Size

Kernel size, the spatial extent of a filter in convolutional neural networks (CNNs), is a critical hyperparameter influencing model performance and efficiency. Current research focuses on optimizing kernel size selection, including adaptive methods that dynamically adjust kernel size based on input data characteristics and novel architectures employing oversized kernels to capture long-range dependencies. These advancements aim to improve accuracy and efficiency in various applications, such as time series classification, image deblurring, and medical image super-resolution, particularly within resource-constrained environments like embedded systems and FPGA-based accelerators.

Papers

August 6, 2024

Effect of Kernel Size on CNN-Vision-Transformer-Based Gaze Prediction Using Electroencephalography Data
Chuhui Qiu, Bugao Liang, Matthew L Key
Mixed Effect Brain Electroencephalography Gaze Prediction Electroencephalogram Data Kernel Size

April 28, 2024

AdaFSNet: Time Series Classification Based on Convolutional Network with a Adaptive and Effective Kernel Size Configuration
Haoxiao Wang, Bo Peng, Jianhua Zhang, Xu Cheng
Time Series Deep Convolutional Neural Network Adaptive Importance Time Series Classification Kernel Size

April 2, 2024

A Methodology for Improving Accuracy of Embedded Spiking Neural Networks through Kernel Size Scaling
Rachmad Vidya Wicaksana Putra, Muhammad Shafique
NCD Method Spiking Neural Network Enhancing Accuracy Accuracy Tradeoff SNN Architecture Embedded Neural Network Kernel Size

February 22, 2024

An FPGA-Based Accelerator Enabling Efficient Support for CNNs with Arbitrary Kernel Sizes
Miaoxin Wang, Xiao Wu, Jun Lin, Zhongfeng Wang
Convolutional Neural Network Hardware Accelerator Inference Accelerator Large Kernel Convolution Kernel Convolutional Neural Network Kernel Size

December 9, 2023

The Counterattack of CNNs in Self-Supervised Learning: Larger Kernel Size might be All You Need
Tianjin Huang, Tianlong Chen, Zhangyang Wang, Shiwei Liu
Convolutional Neural Network Vision Transformer Self Supervised Learning Pre Trained Kernel Size

November 14, 2022

ParCNetV2: Oversized Kernel with Enhanced Attention
Ruihan Xu, Haokui Zhang, Wenze Hu, Shiliang Zhang, Xiaoyu Wang
Convolutional Neural Network Convolutional Block Convolution Kernel Attention Operation Circular Convolution Kernel Size

August 18, 2022

Blind Image Deblurring with Unknown Kernel Size and Substantial Noise
Zhong Zhuang, Taihui Li, Hengkang Wang, Ju Sun
Deep Learning Industrial Disturbing Noise Blind Image Kernel Size

June 20, 2022

Limitations of the NTK for Understanding Generalization in Deep Learning
Nikhil Vyas, Yamini Bansal, Preetum Nakkiran
Deep Learning Strong Generalization Neural Tangent Kernel Fundamental Limitation Kernel Size Layer NTK Scaling Deep Network

April 8, 2022

Multimodal Multi-Head Convolutional Attention with Various Kernel Sizes for Medical Image Super-Resolution
Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Andreea-Iuliana Miron, Olivian Savencu, Nicolae-Catalin Ristea, Nicolae Verga, Fahad Shahbaz Khan
Attention Mechanism Multimodal Phenomenon Channel Attention Attention Head Convolutional Attention Medical Image Super Resolution Kernel Size Multi Head Cross Attention

Kernel Size

Papers

Effect of Kernel Size on CNN-Vision-Transformer-Based Gaze Prediction Using Electroencephalography Data

AdaFSNet: Time Series Classification Based on Convolutional Network with a Adaptive and Effective Kernel Size Configuration

A Methodology for Improving Accuracy of Embedded Spiking Neural Networks through Kernel Size Scaling

An FPGA-Based Accelerator Enabling Efficient Support for CNNs with Arbitrary Kernel Sizes

The Counterattack of CNNs in Self-Supervised Learning: Larger Kernel Size might be All You Need

ParCNetV2: Oversized Kernel with Enhanced Attention

Blind Image Deblurring with Unknown Kernel Size and Substantial Noise

Limitations of the NTK for Understanding Generalization in Deep Learning

Multimodal Multi-Head Convolutional Attention with Various Kernel Sizes for Medical Image Super-Resolution