Order of Compression: A Systematic and Optimal Sequence to Combinationally Compress CNN

TL;DR

This paper addresses the challenge of efficiently compressing CNNs by asking how to optimally sequence multiple compression techniques. It introduces the Order of Compression, analyzes pairwise interactions between Distillation, Pruning, Quantization, and Early Exit, and derives an optimal multi-method sequence using topological sorting. The resulting DPQE order—Distillation (Static Architecture) → Pruning (Static Neuron) → Quantization (Static Sub-Neuron) → Early Exit (Dynamic Architecture)—achieves substantial BitOps reductions (up to $859\times$ on CIFAR-10 with minimal accuracy loss, e.g., $-0.09\%$) and remains robust when adding methods. The work provides a practical, generalizable framework for deploying highly compressed CNNs on resource-limited devices, validated across multiple architectures and datasets.

Abstract

Model compression has gained significant popularity as a means to alleviate the computational and memory demands of machine learning models. Each compression technique leverages unique features to reduce the size of neural networks. Although intuitively combining different techniques may enhance compression effectiveness, we find that the order in which they are combined significantly influences performance. To identify the optimal sequence for compressing neural networks, we propose the Order of Compression, a systematic and optimal sequence to apply multiple compression techniques in the most effective order. We start by building the foundations of the orders between any two compression approaches and then demonstrate inserting additional compression between any two compressions will not break the order of the two compression approaches. Based on the foundations, an optimal order is obtained with topological sorting. Validated on image-based regression and classification networks across different datasets, our proposed Order of Compression significantly reduces computational costs by up to 859 times on ResNet34, with negligible accuracy loss (-0.09% for CIFAR10) compared to the baseline model. We believe our simple yet effective exploration of the order of compression will shed light on the practice of model compression.

Figures (6)

• Figure 1: The Order of Compression Matters: Distillation then pruning will outperforms pruning then distillation or single pruning or single distillation.
• Figure 2: Significant difference over BitOpsCR-Accuracy trade-off with different compression order.
• Figure 3: The established sequence remains unaffected by inserting one more compression approach in the middle.
• Figure 4: Combinational compression with previously established two approaches.
• Figure 5: Repeating compressions.