Sequential Changepoint Detection in Neural Networks with Checkpoints
This addresses the problem of adapting neural networks to changing data distributions in continual learning scenarios, though it appears incremental compared to existing changepoint detection methods.
The paper tackles the problem of online changepoint detection in neural networks during continual learning with unknown task changes, introducing a framework that uses checkpoints and sequential likelihood ratio tests to detect distributional shifts. The result shows improved performance over online Bayesian changepoint detection in challenging applications.
We introduce a framework for online changepoint detection and simultaneous model learning which is applicable to highly parametrized models, such as deep neural networks. It is based on detecting changepoints across time by sequentially performing generalized likelihood ratio tests that require only evaluations of simple prediction score functions. This procedure makes use of checkpoints, consisting of early versions of the actual model parameters, that allow to detect distributional changes by performing predictions on future data. We define an algorithm that bounds the Type I error in the sequential testing procedure. We demonstrate the efficiency of our method in challenging continual learning applications with unknown task changepoints, and show improved performance compared to online Bayesian changepoint detection.