Machine learning, especially deep learning, algorithms have become very popular due to their success in several fields. However, the machine learning model is often treated as a black box. Its understanding, with an explanation about situations of success and failure, may lead to improvements in the design of deep neural networks. Visual analytics can provide visualization, analysis, and user-interaction techniques to include experts in the machine learning loop. In this interactive machine learning paradigm, a question naturally arises: Can visual analytics improve machine learning? We address this question in the context of image-based decision making systems. We present a framework for interactive machine learning of image features and image classifiers, explain the role of each component in a deep neural network, and discuss when and how visual analytics can improve machine learning, including the challenges and research opportunities.
Esta Palestra pode ser assistida ao vivo através de: