Human gaze can serve as an important cue for various application scenarios including context-aware computing and attentive user interfaces. We have been developing a machine learning-based gaze estimation method using large-scale training datasets of face images. Unlike conventional methods that require dedicated hardware, it allows us to use ordinary cameras as gaze estimation device and can be applied to a wider range of applications.

In designing computer vision and machine learning-based systems, there are many cases that it is not enough to consider only the application of trained models. It is crucial to have an interactive framework for letting users create their own recognition models. To make machine learning a democratized tool for general users, we investigate visualization and interaction techniques for interactive machine learning. We are challenging this issue through system development, user evaluation, and workshop studies.