This video presents our representation-centric active learning tool, designed to support closer collaboration between humans and machine learning models. Unlike traditional active learning approaches that rely solely on model-driven query strategies, our system exposes the model’s representation space and enables users to interact directly with its internal representation.

Users can explore the representation space, listen to audio samples, provide annotations, and strategically guide sample selection. As new labels are added, the model retrains and the representation space updates accordingly, supporting a co-evolving understanding between human intuition and model decision-making.

 The project code can be accessed at the link: https://github.com/RidaSaghir/Active-Learning-through-Interaction-with-Model-Representations