Modifying Memories in Transformer Models

TL;DR

This paper introduces a new task for Transformer models to selectively update or erase specific factual knowledge without affecting overall performance, addressing issues like stale information, privacy, and bias.

Contribution

It proposes the task of explicit knowledge modification in Transformers, benchmarks baseline methods, and analyzes key components and training phases affecting knowledge editing.

Findings

Baseline approaches achieve natural performance on the task.

Certain model components are especially effective for knowledge modification.

Training phases influence the model's ability to memorize and modify facts.

Abstract

Large Transformer models have achieved impressive performance in many natural language tasks. In particular, Transformer based language models have been shown to have great capabilities in encoding factual knowledge in their vast amount of parameters. While the tasks of improving the memorization and generalization of Transformers have been widely studied, it is not well known how to make transformers forget specific old facts and memorize new ones. In this paper, we propose a new task of \emph{explicitly modifying specific factual knowledge in Transformer models while ensuring the model performance does not degrade on the unmodified facts}. This task is useful in many scenarios, such as updating stale knowledge, protecting privacy, and eliminating unintended biases stored in the models. We benchmarked several approaches that provide natural baseline performances on this task. This leads to the discovery of key components of a Transformer model that are especially effective for knowledge modifications. The work also provides insights into the role that different training phases (such as pretraining and fine-tuning) play towards memorization and knowledge modification.