Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Masks
This work addresses the challenge of improving neural networks through modularity for researchers and practitioners, but it is incremental as it builds on existing inspection methods.
The paper tackled the problem of understanding functional modularity in neural networks by developing a method using binary weight masks to identify weights and subnets responsible for specific functions, and found that common neural networks fail to reuse submodules, offering insights into systematic generalization in language tasks.
Neural networks (NNs) whose subnetworks implement reusable functions are expected to offer numerous advantages, including compositionality through efficient recombination of functional building blocks, interpretability, preventing catastrophic interference, etc. Understanding if and how NNs are modular could provide insights into how to improve them. Current inspection methods, however, fail to link modules to their functionality. In this paper, we present a novel method based on learning binary weight masks to identify individual weights and subnets responsible for specific functions. Using this powerful tool, we contribute an extensive study of emerging modularity in NNs that covers several standard architectures and datasets. We demonstrate how common NNs fail to reuse submodules and offer new insights into the related issue of systematic generalization on language tasks.