In such cases, the pretrained Transformer block can be skipped.

While deep learning (DL) models have demonstrated impressive performance over a range of complicated tasksanddatamodalities, ithasremained difficulttoquantify theuncertainty fortheir predictions.

GloVE [25] is a 300-dimensional word embedding space. It is an dimensionality-representation representation of word-word co-occurrence statistics. The FlairNLP repository which we used for the preceding represenations can be found here: https://github.com/flairNLP/ Representations were built using a sliding window of 64 words as a context. Representations were built using a sliding window of 64 words as a context.