Towards Efficient Convolutional Neural Network for Embedded Hardware via Multi-Dimensional Pruning

2026-07-13

AI Summary

This work addresses the challenge of balancing accuracy and efficiency in deploying convolutional neural networks on embedded hardware by proposing TECO, a novel framework that jointly prunes network depth, width, and resolution. TECO introduces a two-stage importance evaluation mechanism that integrates intra-dimensional local importance with cross-dimensional global importance, coupled with a heuristic multidimensional pruning algorithm to achieve efficient model compression. Experimental results demonstrate that TECO consistently outperforms state-of-the-art methods across multiple benchmarks, significantly improving inference efficiency on embedded devices while preserving high model accuracy.

Abstract

In this paper, we propose TECO, a multi-dimensional pruning framework to collaboratively prune the three dimensions (depth, width, and resolution) of convolutional neural networks (CNNs) for better execution efficiency on embedded hardware. In TECO, we first introduce a two-stage importance evaluation framework, which efficiently and comprehensively evaluates each pruning unit according to both the local importance inside each dimension and the global importance across different dimensions. Based on the evaluation framework, we present a heuristic pruning algorithm to progressively prune the three dimensions of CNNs towards the optimal trade-off between accuracy and efficiency. Experiments on multiple benchmarks validate the advantages of TECO over existing state-of-the-art (SOTA) approaches. The code and pre-trained models are available at https://github.com/ntuliuteam/Teco.

Problem

• convolutional neural network
• embedded hardware
• multi-dimensional pruning
• model efficiency
• neural network compression

Innovation

• multi-dimensional pruning
• convolutional neural networks
• embedded hardware
• importance evaluation
• model compression