Neural Network Module Decomposition and Recomposition

TL;DR

This paper introduces a modularization technique for deep neural networks that decomposes models into interpretable modules without retraining, enabling efficient recomposition for new tasks with high accuracy and compression.

Contribution

It presents a novel weight mask-based modularization method that does not assume network architecture, allowing flexible decomposition and recomposition of DNNs without retraining.

Findings

Achieves high compression ratio and accuracy in decomposed and recomposed models.

Outperforms existing methods by sharing weights between modules.

Applicable to arbitrary DNN architectures without retraining.

Abstract

We propose a modularization method that decomposes a deep neural network (DNN) into small modules from a functionality perspective and recomposes them into a new model for some other task. Decomposed modules are expected to have the advantages of interpretability and verifiability due to their small size. In contrast to existing studies based on reusing models that involve retraining, such as a transfer learning model, the proposed method does not require retraining and has wide applicability as it can be easily combined with existing functional modules. The proposed method extracts modules using weight masks and can be applied to arbitrary DNNs. Unlike existing studies, it requires no assumption about the network architecture. To extract modules, we designed a learning method and a loss function to maximize shared weights among modules. As a result, the extracted modules can be recomposed without a large increase in the size. We demonstrate that the proposed method can decompose and recompose DNNs with high compression ratio and high accuracy and is superior to the existing method through sharing weights between modules.