Adaptive Logistic Regression: Leveraging Arctan in Active Learning

Creating labeled data for supervised learning can frequently be expensive and time-consuming, which makes it difficult to train efficient models. By selectively querying the most informative data points for labeling, active learning provides a strategy that lowers the quantity of labeled data needed while preserving model performance. In this work, a method for incorporating the Arctan function into logistic regression for active learning scenarios is presented. The logistic regression model's stability and resilience are improved by utilizing the Arctan function, which has a smooth and gentle transition between probability limits, especially when working with sparsely labeled data. Our approach aims to improve the efficiency of model training in situations where labeled data is scarce. Through a series of simulations, we demonstrate that the Arctan-based logistic regression model also achieves high accuracy in terms of convergence speed and generalization capability. The results indicate that this method is a promising alternative for applications where data labeling is a bottleneck, offering a more efficient and effective solution for supervised learning tasks. This research contributes to the ongoing exploration of alternative activation functions in machine learning and their potential to improve model performance in challenging scenarios.