Rank-One Editing of Encoder-Decoder Models
This addresses the need for efficient model updates in real-world deployments, though it is incremental as it builds on existing editing techniques.
The paper tackles the problem of adapting encoder-decoder models for behavior deletion requests without retraining, proposing rank-one editing as a direct intervention method that achieves high efficacy using only a single positive example to fix erroneous behaviors in Neural Machine Translation.
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.