Deep neural networks are not good at telling when they are not sure while working in an active learning environment. They have grown so complex that it seems impossible to follow their decision-making processes. However, we have little control over rebuilding it as it is not transparent to us.
Can we explain how networks come to decisions by imitating human-like reasoning in multiple-choice testing?
This study looks into an active learning environment in an educational setting. A new pool-based query strategy measures the information capacity of data using the information function from the four-parameter logistic item response theory (4PL IRT).
We compared the proposed strategy with the most common active learning strategies—Least Confidence and Entropy Sampling. We provide a notebook to illustrate Information Capcity on MNIST and Fashion MNIST datasets: Human-like active learning.ipynb
@Article{e22080906,
AUTHOR = {Kulikovskikh, Ilona and Lipic, Tomislav and Šmuc, Tomislav},
TITLE = {From Knowledge Transmission to Knowledge Construction: A Step towards Human-Like Active Learning},
JOURNAL = {Entropy},
VOLUME = {22},
YEAR = {2020},
NUMBER = {8},
ARTICLE-NUMBER = {906},
URL = {https://www.mdpi.com/1099-4300/22/8/906},
DOI = {10.3390/e22080906}
}
Please contact Ilona Kulikovskikh at kulikovskikh.im@ssau.ru for questions.