STENCIL: Submodular Mutual Information Based Weak Supervision for Cold-Start Active Learning

arXiv:2402.13468v1 Announce Type: new
Abstract: As supervised fine-tuning of pre-trained models within NLP applications increases in popularity, larger corpora of annotated data are required, especially with increasing parameter counts in large language models. Active learning, which attempts to mine and annotate unlabeled instances to improve model performance maximally fast, is a common choice for reducing the annotation cost; however, most methods typically ignore class imbalance and either assume access to initial annotated data or require multiple rounds of active learning …

abstract active learning annotated data applications arxiv cs.cl cs.lg data fine-tuning information instances language language models large language large language models mine nlp performance pre-trained models supervised fine-tuning supervision type