Low-Complexity Probing via Finding Subnetworks

TL;DR

This paper introduces a subtractive pruning-based subnetwork probing method that outperforms traditional MLP probes in detecting linguistic properties in neural networks, with better accuracy and lower complexity.

Contribution

It proposes a novel subnetwork probing technique that finds existing subnetworks encoding linguistic properties, reducing the need for additional parameters and improving detection accuracy.

Findings

Subnetworks outperform MLP probes in accuracy and complexity.

Lower-level tasks are encoded in lower layers of neural networks.

Subnetwork probing better distinguishes properties from random models.

Abstract

The dominant approach in probing neural networks for linguistic properties is to train a new shallow multi-layer perceptron (MLP) on top of the model's internal representations. This approach can detect properties encoded in the model, but at the cost of adding new parameters that may learn the task directly. We instead propose a subtractive pruning-based probe, where we find an existing subnetwork that performs the linguistic task of interest. Compared to an MLP, the subnetwork probe achieves both higher accuracy on pre-trained models and lower accuracy on random models, so it is both better at finding properties of interest and worse at learning on its own. Next, by varying the complexity of each probe, we show that subnetwork probing Pareto-dominates MLP probing in that it achieves higher accuracy given any budget of probe complexity. Finally, we analyze the resulting subnetworks across various tasks to locate where each task is encoded, and we find that lower-level tasks are captured in lower layers, reproducing similar findings in past work.