Model Compression for Domain Adaptation through Causal Effect Estimation

TL;DR

This paper introduces a causal effect estimation approach to model compression aimed at improving domain adaptation in NLP, by selecting model components that maximize out-of-distribution performance.

Contribution

It proposes the ATE-guided Model Compression (AMoC) method, which leverages causal effects of model components to optimize compressed models for domain transfer.

Findings

AMoC outperforms strong baselines across multiple tasks.

The method effectively identifies model components critical for domain generalization.

Causal effect estimation improves model compression strategies for domain adaptation.

Abstract

Recent improvements in the predictive quality of natural language processing systems are often dependent on a substantial increase in the number of model parameters. This has led to various attempts of compressing such models, but existing methods have not considered the differences in the predictive power of various model components or in the generalizability of the compressed models. To understand the connection between model compression and out-of-distribution generalization, we define the task of compressing language representation models such that they perform best in a domain adaptation setting. We choose to address this problem from a causal perspective, attempting to estimate the average treatment effect (ATE) of a model component, such as a single layer, on the model's predictions. Our proposed ATE-guided Model Compression scheme (AMoC), generates many model candidates, differing by the model components that were removed. Then, we select the best candidate through a stepwise regression model that utilizes the ATE to predict the expected performance on the target domain. AMoC outperforms strong baselines on dozens of domain pairs across three text classification and sequence tagging tasks.