Towards Robust Pruning: An Adaptive Knowledge-Retention Pruning Strategy for Language Models
This work addresses the challenge of robust pruning for language models, which is crucial for deploying efficient models in adversarial environments, though it appears incremental as it builds on existing pruning objectives.
The paper tackled the problem of maintaining robustness in pruned language models against adversarial attacks as sparsity increases, proposing an adaptive knowledge-retention pruning strategy that replicates embedding and feature spaces to conserve pre-trained knowledge, resulting in a superior balance of accuracy, sparsity, robustness, and pruning cost on BERT across datasets like SST2, IMDB, and AGNews.
The pruning objective has recently extended beyond accuracy and sparsity to robustness in language models. Despite this, existing methods struggle to enhance robustness against adversarial attacks when continually increasing model sparsity and require a retraining process. As humans step into the era of large language models, these issues become increasingly prominent. This paper proposes that the robustness of language models is proportional to the extent of pre-trained knowledge they encompass. Accordingly, we introduce a post-training pruning strategy designed to faithfully replicate the embedding space and feature space of dense language models, aiming to conserve more pre-trained knowledge during the pruning process. In this setup, each layer's reconstruction error not only originates from itself but also includes cumulative error from preceding layers, followed by an adaptive rectification. Compared to other state-of-art baselines, our approach demonstrates a superior balance between accuracy, sparsity, robustness, and pruning cost with BERT on datasets SST2, IMDB, and AGNews, marking a significant stride towards robust pruning in language models.