LSTM Ensemble¶
Implementation of an ensemble of nlp_uncertainty_zoo.models.lstm instances.
This is inspired by the work of Lakshminarayanan et al. (2017), showing that an ensemble is a strong model for generalization and uncertainty quantification due to its diversity.
This was also confirmed in the work of Ovadia et al. (2019) in a computer vision setting, and in Ulmer et al. (2022) for natural language processing, where ensembles for LSTMs performed en par or better than pre-trained BERT models.