Active Learning: Encoder-Decoder-Outlayer and Vector Space Diversification Sampling

This study introduces a training pipeline comprising two components: the Encoder-Decoder-Outlayer framework and Vector Space Diversification Sampling method. efficiently separates pre-training fine-tuning stages, while sampling method employs pivot nodes to divide subvector space selectively choose unlabeled data, thereby reducing reliance on human labeling. The offers numerous advantages, including rapid training, parallelization, buffer capability, flexibility, low GPU memory usage, sample with nearly linear time complexity. Experimental results demonstrate that models trained proposed algorithm generally outperform those random small datasets. These characteristics make it highly efficient effective approach for machine learning models. Further details can be found in project repository GitHub.