Mass-Editing Memory in a Transformer

By Kevin Meng, Arnab Sen Sharma, Alex Andonian, Yonatan Belinkov, David Bau. 2022

1,108 citations recorded by the source corpus.

Recent work has shown exciting promise in updating large language models with new memories, so as to replace obsolete information or add specialized knowledge. However, this line of work is predominantly limited to updating single associations. We develop MEMIT, a method for directly updating a language model with many memories, demonstrating experimentally that it can scale up to thousands of associations for GPT-J (6B) and GPT-NeoX (20B), exceeding prior work by orders of magnitude. Our code and data are at https://memit.baulab.info.

Subfields: Artificial Intelligence, Computer Networks and Communications, Information Systems and Management, Large Language Models, 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 Computer Networks and Communications

In Information Systems and Management

In Large Language Models

In Safety, Robustness & Interpretability

Citations in the corpus

Locating and Editing Factual Associations in GPT (2022, methodology): “Adapting the approach to scale up to many more facts is the subject of other work such as Meng, Sen Sharma, Andonian, Belinkov, and Bau (2022).”
Can We Edit Factual Knowledge by In-Context Learning? (2023, background): “E.1 Scale up to more factual edits Mitchell et al. (2022b); Meng et al. (2022b) find that gradient-based knowledge editing methods encounter difficulties when attempting to update multiple knowledge facts simultaneously.”
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023, positive): “MEMIT (Meng et al., 2023) employs the same knowledge locating methods as ROME, enabling the concurrent updating of multiple layers to facilitate the simultaneous integration of thousands of editing knowledge.”
Unified Concept Editing in Diffusion Models (2023, background): “Except for this watermark, it is identical to the accepted version; the final published version of the proceedings is available on IEEE Xplore. thesis model while preserving the generative quality of the model for unedited concepts.”
Unifying Large Language Models and Knowledge Graphs: A Roadmap (2023, background): “Existing studies [252], [253] also propose solutions to edit knowledge in LLMs, but only limit themselves to handling simple tuple-based knowledge in KGs.”
Editing Large Language Models: Problems, Methods, and Opportunities (2023, methodology): “KN For Knowledge Neuron (Dai et al., 2022), we follow Meng et al. (2023) re-implementation in their study.”
Improving Alignment and Robustness with Circuit Breakers (2024, neutral): “This includes research ranging from uncovering emergent interpretable structures in intermediate representations [77, 46, 10], to the identification and modification of embedded knowledge [48, 42, 43], as well as steering model outputs [66, 7, 33, 24, 65].”
Dissecting Recall of Factual Associations in Auto-Regressive Language Models (2023, background): “A prominent thread focused on locating layers (Meng et al., 2022a; Wallat et al., 2020) and neurons (Dai et al., 2022) that store factual information, which often informs editing approaches (De Cao et al., 2021; Mitchell et al., 2022; Meng et al., 2022b).”
MACE: Mass Concept Erasure in Diffusion Models (2024, methodology): “…methods that view matrices as linear associative memories [1, 28], often used to edit knowledge embedded within neural networks [2, 4, 5, 17, 36, 37, 43], we formulate our objective function as follows: where λ 1 ∈ R + is a hyperparameter, e is the embedding of a word co-existing with the…”
MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions (2023, methodology): “In addition, they all gain traction on edit-wise evaluation, with MEMIT and ROME achieving Table 4: Performance results on MQ U AKE-T for different knowledge editing methods using GPT-J as the base model.”
Natural Language Processing with Commonsense Knowledge: A Survey (2021, methodology): “Compared to the vanilla MEMIT method [108], MEMIT CSK improves it for the commonsense domain by varying edit tokens and improving the layer selection strategy.”
Continual Learning of Large Language Models: A Comprehensive Survey (2024, positive): “This challenges the classical “locate and edit” paradigm used by several existing methods [157, 158], and could become a significent concern for CMR [85].”
Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, positive): “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., 2022a,b).”
Ablating Concepts in Text-to-Image Diffusion Models (2023, background): “Similarly, model editing [5, 70, 19, 69, 45, 38, 40, 39] aims to modify specific model weights based on users’ instructions to incorporate new computational rules or new visual effects.”
Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptors (2022, neutral): “…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].”
Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback (2023, methodology): “In robotics, correction-based feedback has been used for improving policies [Li et al., 2021, Losey et al., 2022, Bajcsy et al., 2018] and plans [Sharma et al., 2022].”