Rank-One Editing of Encoder-Decoder Models
Preprint
Abstract
Large sequence to sequence models for tasks such as Neural Machine Translation (NMT) are usually trained over hundreds of millions of samples. However, training is just the origin of a model's life-cycle. Real-world deployments of models require further behavioral adaptations as new requirements emerge or shortcomings become known. Typically, in the space of model behaviors, behavior deletion requests are addressed through model retrainings whereas model finetuning is done to address behavior addition requests, both procedures being instances of data-based model intervention. In this work, we present a preliminary study investigating rank-one editing as a direct intervention method for behavior deletion requests in encoder-decoder transformer models. We propose four editing tasks for NMT and show that the proposed editing algorithm achieves high efficacy, while requiring only a single instance of positive example to fix an erroneous (negative) model behavior.Keywords
All Related Versions
This publication has 7 references indexed in Scilit:
- Dataset Security for Machine Learning: Data Poisoning, Backdoor Attacks, and DefensesIEEE Transactions on Pattern Analysis and Machine Intelligence, 2022
- Transformer Feed-Forward Layers Are Key-Value Memories2021
- The Curious Case of Hallucinations in Neural Machine Translation2021
- Rewriting a Deep Generative Model2020
- Continual Lifelong Learning in Natural Language Processing: A Survey2020
- Beyond Accuracy: Behavioral Testing of NLP Models with CheckList2020
- Overcoming Catastrophic Forgetting During Domain Adaptation of Neural Machine Translation2019