Active Learning in CNNs via Expected Improvement Maximization

Udai G. Nagpal, David A. Knowles

Keywords: active learning, batch-mode active learning, deep learning, convolutional neural networks, supervised learning, regression, classification, MC dropout, computer vision, computational biology

Abstract: Deep learning models such as Convolutional Neural Networks (CNNs) have demonstrated high levels of effectiveness in a variety of domains, including computer vision and more recently, computational biology. However, training effective models often requires assembling and/or labeling large datasets, which may be prohibitively time-consuming or costly. Pool-based active learning techniques have the potential to mitigate these issues, leveraging models trained on limited data to selectively query unlabeled data points from a pool in an attempt to expedite the learning process. Here we present "Dropout-based Expected IMprOvementS" (DEIMOS), a flexible and computationally-efficient approach to active learning that queries points that are expected to maximize the model's improvement across a representative sample of points. The proposed framework enables us to maintain a prediction covariance matrix capturing model uncertainty, and to dynamically update this matrix in order to generate diverse batches of points in the batch-mode setting. Our active learning results demonstrate that DEIMOS outperforms several existing baselines across multiple regression and classification tasks taken from computer vision and genomics.

One-sentence Summary: An efficient batch-mode active learning algorithm for CNNs is proposed based on acquisition of points expected to maximize the model’s improvement upon being queried, and is found to perform well across regression and classification tasks.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2011.14015/code)

Reviewed Version (pdf): https://openreview.net/references/pdf?id=PUGPI7Wlaw