Automated Audio Captioning with Epochal Difficult Captions for curriculum learning

Audio captioning and retrieval with improved cross-modal objectives

“…Epochal Difficult Captions with Curriculum Learning [30] In the 3rd study, we analyse the effectiveness of scheduling data and gradually introducing harder samples for Automated Audio Captioning. This method influences the generation quality and learning of the captioning model by presenting the training captions in a way that builds upon the model's existing knowledge and skills from previous training epochs.…”

“…In the third study published at APSIPA 2022 [30] for Automated Audio Captioning, Epochal Difficult Captions with Curriculum Learning is proposed to analyse the effectiveness of scheduling data and to introduce samples with a difficulty level relative to the epoch. This method improves the generation quality and involves presenting the training captions in a manner that leverages the model's existing knowledge and skills from prior training epochs, inspired by models trained on a curriculum to accomplish mathematical proving [25].…”