Memory-Based Model Editing at Scale

By Eric Mitchell, Charles Lin, Antoine Bosselut, Christopher D. Manning, Chelsea Finn. 2022

561 citations recorded by the source corpus.

Even the largest neural networks make errors, and once-correct predictions can become invalid as the world changes. Model editors make local updates to the behavior of base (pre-trained) models to inject updated knowledge or correct undesirable behaviors. Existing model editors have shown promise, but also suffer from insufficient expressiveness: they struggle to accurately model an edit's intended scope (examples affected by the edit), leading to inaccurate predictions for test inputs loosely related to the edit, and they often fail altogether after many edits. As a higher-capacity alternative, we propose Semi-Parametric Editing with a Retrieval-Augmented Counterfactual Model (SERAC), which stores edits in an explicit memory and learns to reason over them to modulate the base model's predictions as needed. To enable more rigorous evaluation of model editors, we introduce three challenging language model editing problems based on question answering, fact-checking, and dialogue generation. We find that only SERAC achieves high performance on all three problems, consistently outperforming existing approaches to model editing by a significant margin. Code, data, and additional project information will be made available at https://sites.google.com/view/serac-editing.

Subfields: Artificial Intelligence, Model Architectures, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Model Architectures

In Safety, Robustness & Interpretability

Citations in the corpus

GLM-130B: An Open Bilingual Pre-trained Model (2022, methodology): “Researchers can also modify the model architecture and weights, to validate the proposed algorithms to improve LLMs Zhu et al. (2020); Cao et al. (2021); Hase et al. (2021); Mitchell et al. (2022).”
Mass-Editing Memory in a Transformer (2022, methodology): “Achieving these goals is challenging, even for a few edits (Hase et al., 2021; Mitchell et al., 2022; Meng et al., 2022).”
Rethinking machine unlearning for large language models (2024, unknown): “Thus, existing memory-based model editing techniques (Li et al., 2022; Mitchell et al., 2022; Madaan et al., 2022; Zheng et al., 2023) focus on updating short-term working memory instead of altering the long-term memory encapsulated in the model’s weights.”
Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptors (2022, unknown): “…between layers. c) Edits appear sporadically and require quick fixes, so GRACE codebooks are curated over long sequences of edits. rely on regularized finetuning [8, 28, 29, 37] succumb to overfitting [4], while hypernetwork methods require pre-training on hard-to-access data [30, 31].”
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023, unknown): “SERAC (Mitchell et al., 2022b) employs a scope classifier to route the input associated with new knowledge stored in an external edit memory toward the counterfactual model, which can aid the base model in handling the updated information.”
Can We Edit Factual Knowledge by In-Context Learning? (2023, methodology): “Mitchell et al. (2022b) proposes a memory-based retrieval-augmented method to handle multiple factual edits.”
Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, unknown): “The first involves the incorporation of an auxiliary sub-network (Mitchell et al., 2022; Huang et al., 2023b), while the second entails direct modification of the original model parameters (Meng et al.”
Editing Models with Task Arithmetic (2022, unknown): “[70] Eric Mitchell, Charles Lin, Antoine Bosselut, Christopher D Manning, and Chelsea Finn.”
Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning (2024, methodology): “Finally, we note the existence of attack methods for extracting data from unlearned models (Shi et al., 2023; Patil et al., 2023), and other unlearning methods including model editing (Mitchell et al., 2022; Meng et al., 2022) and in-context unlearning (Pawelczyk et al., 2023).”
Reasoning with Language Model Prompting: A Survey (2022, background): “Other promising directions include retrieval augmentation (Li et al., 2022a), model editing (Cao et al., 2021; Mitchell et al., 2022a,b; Cheng et al., 2023), delta-tuning (He et al., 2022; Mao et al., 2022; Pal et al., 2022; Ding et al., 2022), etc. Robust, Faithful and Interpretable Reasoning.”
RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval (2024, unknown): “…knowledge for particular tasks and the world continues to change, invalidating facts in the LLM. Updating the knowledge of these models through additional fine-tuning or editing is difficult, particularly when dealing with vast text corpora (Lewis et al., 2020; Mitchell et al., 2022).”
Editing Large Language Models: Problems, Methods, and Opportunities (2023, unknown): “Model editing, as elucidated by Mitchell et al. (2022b), aims to adjust an initial base model’s ( f θ , θ signifies the model’s parameters) behavior on the particular edit descriptor ( x e , y e ) efficiently without influencing the model behavior on other samples.”
MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions (2023, methodology): “0% for MEND), as instance-wise correctness relies on recalling every fact from the model correctly for multi-hop questions.”