A meta-active learning approach exploiting instance importance

This repository contains the code and resources for the research paper "A meta-active learning approach exploiting instance importance". The Algorithms "LAL-IGrad" and "LAL-IGrad-VAE" are two meta-active learning algorithm that learns to estimate sample difficulty by leveraging neural network gradients.

Citation

@article{FLESCA2024123320,
title = {A meta-active learning approach exploiting instance importance},
journal = {Expert Systems with Applications},
volume = {247},
pages = {123320},
year = {2024},
issn = {0957-4174},
doi = {https://doi.org/10.1016/j.eswa.2024.123320},
url = {https://www.sciencedirect.com/science/article/pii/S0957417424001854},
author = {Sergio Flesca and Domenico Mandaglio and Francesco Scala and Andrea Tagarelli},
keywords = {Active learning, Neural networks, Instance importance, Meta-learning, Classification, Gradient variation}
}