An active learning framework to optimize training of deep models with human in the loop

Description

Deep learning models have been used extensively to solve real-world problems, but the performance of such models relies heavily on large amounts of labeled data for training. While advances in data collection technology have enabled the acquisition of a massive volume of data, labeling the data remains an expensive and time-consuming task. The random selection of data points for generating training data is not always useful, particularly when there’s a high variation in the scale, shape, and orientation of the data or when sample data points do not follow an even distribution. However, active learning techniques are being progressively adopted to accelerate the development of machine learning solutions by allowing the model to query the data they learn from.

Humayun Irshad offers an overview of an active learning framework that uses a crowdsourcing approach to solve parking sign recognition—a real-world problem in transportation and autonomous driving for which a large amount of unlabeled data is available. The main novelty of the framework is a crowdsourcing-based active learning framework which intelligently selects a small subset of images that efficiently improve the performance of the object detection model. For the iterative part of the proposed active learning framework, Humayun defined a criteria for the selection of images that are good candidates for building a better model. These candidate images were chosen to cover not only a diverse range of parking signs but also challenging corner cases where the parking signs are either partially occluded, blurred, reflecting light, or are placed far away in the background. Join in to learn how the solution generates an accurate model, quickly and cost-effectively, despite the unevenness of the data.