Convolution Primitive

Convolution primitives are fundamental building blocks of convolutional neural networks (CNNs), and optimizing their performance is crucial for efficient deep learning. Current research focuses on developing and benchmarking alternative algorithms to the commonly used General Matrix Multiplication (GEMM) approach, including sliding window techniques and specialized kernels tailored to specific hardware architectures like ARM Cortex-M processors. These efforts aim to reduce computational cost and memory usage, enabling faster and more energy-efficient CNN inference, particularly on resource-constrained devices. The resulting improvements have significant implications for deploying AI in embedded systems and expanding the accessibility of deep learning applications.

Papers

July 15, 2024

ConvBench: A Comprehensive Benchmark for 2D Convolution Primitive Evaluation
Lucas Alvarenga, Victor Ferrari, Rafael Souza, Marcio Pereira, Guido Araujo
Comprehensive Benchmark Convolution Operation Learning Benchmark Convolution Technique Convolution Head Convolution Primitive

October 8, 2023

Accelerating Machine Learning Primitives on Commodity Hardware
Roman Snytsar
Sliding Window Computation Kernel Convolution Primitive

May 25, 2023

Sliding Window Sum Algorithms for Deep Neural Networks
Roman Snytsar
Deep Neural Network Convolution Kernel Memory Management Computation Kernel Convolution Primitive

March 19, 2023

Evaluation of Convolution Primitives for Embedded Neural Networks on 32-bit Microcontrollers
Baptiste Nguyen, Pierre-Alain Moellic, Sylvain Blayac
Global Evaluation 32 Bit Microcontrollers Traditional Convolution Embedded Neural Network Shift Convolution Convolution Primitive

Convolution Primitive

Papers

ConvBench: A Comprehensive Benchmark for 2D Convolution Primitive Evaluation

Accelerating Machine Learning Primitives on Commodity Hardware

Sliding Window Sum Algorithms for Deep Neural Networks

Evaluation of Convolution Primitives for Embedded Neural Networks on 32-bit Microcontrollers