Active Learning as a Way of Increasing Accuracy

Abstract

Abstract-In machine-learning areas, number of the data for training process alters the success of models. More samples in training give more success. However obtaining data with label information is costly and long-lasting process. Active learning algorithms are emerged to overcome this problem. It can be used with any machine learning algorithms. Active learning algorithms try to maintain same success resulted by regular machine learning methods with fewer samples. In this study, a modified active learning algorithm tested on six datasets with different machine learning methods. Comparative results presented with charts in result. Algorithm are not only providing same success but also slightly increasing total success with smarter training process.Index Terms-Active learning, random forest, single vector machines, k-nearest neighbor, naï ve bayes, machine learning.

Improving Deep Learning-Based Defect Classification in Solar Cells Using Conformal Prediction

“…Another important framework in machine learning is the concept of active learning, which can be used to define an algorithm driven training loop that relies on a machine learning model and some measure of information content that can be used to ask a human expert for input on new instances that are of most value to further improve the machine learning model. During active learning, the machine learning model is then conveniently retrained to be updated using newly labeled data, allowing for optimal model improvement over time while keeping the cost of labeling data to a minimum as only a minimal amount of new images is seen by an expert for labeling [7] [8]. Active learning can be done in several ways, and in this work conformal prediction is used to define an uncertaintyaware active learning loop for determining the low-confidence data to be labeled by a human.…”

Multiclass Deep Active Learning for Detecting Red Blood Cell Subtypes in Brightfield Microscopy

“…One promising approach to break the bottleneck of data annotation is active learning, which uses a learning algorithm that carefully selects unlabeled data points to interactively query experts for new annotations. This expert-in-the-loop process has been demonstrated to achieve similar or even greater performance as compared to a fully labelled data-set, with a fraction of the cost and time that it takes to label all the data [9]. Here, we combine active learning with object detection and develop a novel active learning annotation tool to guide expert annotation.…”