Editable Neural Networks

By Anton Sinitsin, Vsevolod Plokhotnyuk, Dmitriy V. Pyrkin, Sergei Popov, Artem Babenko. 2020

220 citations recorded by the source corpus.

These days deep neural networks are ubiquitously used in a wide range of tasks, from image classification and machine translation to face identification and self-driving cars. In many applications, a single model error can lead to devastating financial, reputational and even life-threatening consequences. Therefore, it is crucially important to correct model mistakes quickly as they appear. In this work, we investigate the problem of neural network editing - how one can efficiently patch a mistake of the model on a particular sample, without influencing the model behavior on other samples. Namely, we propose Editable Training, a model-agnostic training technique that encourages fast editing of the trained model. We empirically demonstrate the effectiveness of this method on large-scale image classification and machine translation tasks.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing

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 Vision and Pattern Recognition

In Natural Language Processing

Citations in the corpus

Time-Aware Language Models as Temporal Knowledge Bases (2021, background): “A related line of work has explored editing neural predictions after training given a dataset of revised input and output pairs (Sinitsin et al., 2020; Zhu et al., 2020; De Cao et al., 2021).”
Mass-Editing Memory in a Transformer (2022, background): “Sinitsin et al. (2019) proposes a training objective to produce models amenable to editing by gradient descent.”
Do Language Models Have Beliefs? Methods for Detecting, Updating, and Visualizing Model Beliefs (2021, methodology): “We build on work on editing models after training, which is an exciting recent direction of research with many potentially valuable use cases (Sinitsin et al., 2020; Zhu et al., 2020; De Cao et al., 2021; Mitchell et al., 2021).”
Fast Model Editing at Scale (2021, methodology): “However, a key challenge in deploying and maintaining such models is issuing patches to adjust model behavior after deployment (Sinitsin et al., 2020).”
Editing Factual Knowledge in Language Models (2021, methodology): “Sinitsin et al. (2020) measure performance deterioration and success rate but do not measure retain accuracy nor equivalence accuracy .”
Meta-Learning in Neural Networks: A Survey (2020, background): “A related idea is meta-training representations to support local editing updates [259] for improvementwithout interference.”
Mind the Gap: Assessing Temporal Generalization in Neural Language Models (2021, background): “A related approach is to disentangle the acquisition of up-to-date knowledge from the language learning itself, and enable direct editing of that knowledge within the model (Sinitsin et al., 2020; Zhu et al., 2020; De Cao et al., 2021).”
Memory-Based Model Editing at Scale (2022, methodology): “As points of comparison, we consider gradient-based editors, including ﬁne-tuning on the edit example ( FT ), editable neural networks ( ENN ; Sinitsin et al., 2020), model editor networks using gradient decomposition ( MEND ; Mitchell et al., 2021), as well as a cache+lookup baseline LU 4 .”
Editing Large Language Models: Problems, Methods, and Opportunities (2023, background): “As shown in Figure 1, to address this issue, the concept of model editing has been proposed (Sinitsin et al., 2020; De Cao et al., 2021), enabling data-efficient alterations to the behavior of models, specifically within a designated realm of interest, while ensuring no adverse impact on other…”
Entity Cloze By Date: What LMs Know About Unseen Entities (2022, background): “Since the entity knowledge in our dataset is time-indexed, this is suitable for evaluating knowledge editing approaches (Sinitsin et al., 2020; Zhu et al., 2020; De Cao et al., 2021; Mitchell et al., 2021; Meng et al., 2022) and also continual knowledge learning approaches (Jang et al., 2021).”
Provable repair of deep neural networks (2021, unknown): “This dataset was also used by Sinitsin et al. [53].”
Post-hoc Concept Bottleneck Models (2022, background): “One thread of work focuses on intervening on the latent space of neural networks to alter the generated output towards the desired state, e.g. removal of artifacts or manipulation of object positions (Sinitsin et al., 2020; Bau et al., 2020; Santurkar et al., 2021).”
Knowledge Conflicts for LLMs: A Survey (2024, unknown): “Prior works tackle temporal misalignment by focusing on three lines of strategies: Knowledge editing (KE) aims to directly update the parametric knowledge (Sinitsin et al., 2020; Mitchell et al., 2021; Onoe et al., 2023).”
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023, unknown): “One is Knowledge editing (Sinitsin et al., 2020; Yao et al., 2023c), which aims at directly editing model parameters to bridge the knowledge gap.”
Aging with GRACE: Lifelong Model Editing with Discrete Key-Value Adaptors (2022, unknown): “Most methods propose variants of regularized-finetuning via auxiliary data, like training instances from the pre-trained model’s training data or by using semantically-equivalent versions of new edits [37].”
Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Models (2023, unknown): “To mitigate these issues in LLMs with minimal computational overhead, the concept of model editing has been introduced (Sinitsin et al., 2020; De Cao et al., 2021).”
Continual Learning of Large Language Models: A Comprehensive Survey (2024, unknown): “For more detailed introduction, please refer to Appendix A.1.”
DialGuide: Aligning Dialogue Model Behavior with Developer Guidelines (2022, methodology): “Fixing Models through Intervention Some recent work has explored editing models by computing targeted changes in the model’s parameters (Sinitsin et al., 2020; Hase et al., 2021; Mitchell et al., 2021; Meng et al., 2022), while others have explored natural language feedback (Madaan et al., 2022;…”
MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions (2023, background): “As large language models (LLMs) are deployed widely, the need to keep their knowledge correct and up-to-date without massive retraining costs becomes increasingly important (Sinitsin et al., 2020).”