Active learning (AL) is a machine learning technique that aims to reduce annotation costs by selectively choosing the most informative samples for labeling. This process relies on acquisition functions, which can be broadly categorized into two types: representativity-based and uncertainty-based. Representativity-based functions focus on exploring the dataset, while uncertainty-based functions refine decision boundaries. This creates a trade-off known as the exploration-exploitation dilemma. To address this challenge, we propose a novel approach that alternates between these two types of acquisition functions. Our method employs an adaptive feedback-driven selection mechanism, an annealing-based approach, or a baseline random criterion to guide the alternation process. This strategy helps mitigate common AL issues, such as batch mode inefficiency and cold start problems. Our experiments demonstrate that the alternating approach enhances both the accuracy and robustness of the AL process. Additionally, we consider the balance between accuracy and energy consumption, contributing to the development of more sustainable AI systems. By evaluating our criterion across various models and datasets, we show its potential to reduce computational costs while maintaining or even improving accuracy. Notably, alternating between the BALD and BADGE acquisition functions yields particularly robust results.

Active learning (AL) is a machine learning (ML) approach that entails carefully choosing the most informative samples for annotation during training, aiming to minimize annotation costs. AL has recently emerged as a promising approach in the context of green ML, as an energy-efficient learning method on top of being data-efficient. Nevertheless, given the significant cost of AL, it might lead one to question the effectiveness of this approach in reducing computational costs and promoting green ML. In this paper, we conduct a comparative analysis of both fundamental and advanced active learning methods against a random baseline selection, aiming to demonstrate the efficacy of active learning to reduce the cost of training. This study demonstrates that, with careful tuning of hyperparameters like query size and pool size, AL is able to reduce runtime while maintaining competitive accuracy for classification tasks.

Each year, pneumonia causes harm to millions of people worldwide. The chest radiograph remains the most important tool for diagnosing pneumonia and many other thoracic diseases. Many deep-learning approaches have been advocated in recent years, taking advantage of large public medical datasets and powerful computational resources to support radiologists in their examinations. However, the vast amount of data available is rarely utilized. In this work, we propose two novel attention-based mechanisms for fusing multimodal and multi-view data in a medical context, respectively. A unimodal multi-view classifier for chest X-rays and a multimodal classifier are developed, inter alia, operating these mechanisms to classify pneumonia using chest radiographs in combination with other clinically derived modalities. Our unimodal multi-view model outperforms other state-of-the-art approaches on the MIMIC-IV dataset in the AUROC metric by at least 9.2%, achieving an AUROC score of 81.5%. Our attention-based multimodal model, namely Radiopaths, achieved an AUROC score of 87.3%, further increasing the multi-view model's performance by 5.8%, while acquiring robustness to missing modalities and scalability.