Learning sparse and meaningful representations through embodiment.

How do humans acquire a meaningful understanding of the world with little to no supervision or semantic labels provided by the environment? Here we investigate embodiment with a closed loop between action and perception as one key component in this process. We take a close look at the representations learned by a deep reinforcement learning agent that is trained with high-dimensional visual observations collected in a 3D environment with very sparse rewards. We show that this agent learns stable representations of meaningful concepts such as doors without receiving any semantic labels. Our results show that the agent learns to represent the action relevant information, extracted from a simulated camera stream, in a wide variety of sparse activation patterns. The quality of the representations learned shows the strength of embodied learning and its advantages over fully supervised approaches.

A Quantitative Analysis of the Taxonomy of Artistic Styles.

Classifying artists and their work as distinct art styles has been an important task of scholars in the field of art history. Due to its subjectivity, scholars often contradict one another. Our project investigated differences in aesthetic qualities of seven art styles through quantitative means. This was achieved with state-of-the-art deep-learning paradigms to generate new images resembling the style of an artist or entire era. We conducted psychological experiments to measure the behavior of subjects when viewing these new art images. Two different experiments were used: In an eye-tracking study, subjects viewed art-stylespecific generated images. Eye movements were recorded and then compared between art styles. In a visual singleton search study, subjects had to locate a style-outlier image among three images of an alternative style. Reaction time and accuracy were measured and analyzed. These experiments show that there are measurable differences in behavior when viewing images of varying art styles. From these differences, we constructed hierarchical clusterings relating art styles based on the different behaviors of subjects viewing the samples. Our study reveals a novel perspective on the classification of artworks into stylistic eras and motivates future research in the domain of empirical aesthetics through quantitative means.

Learning of Spatial Properties of a Large-Scale Virtual City With an Interactive Map.

To become acquainted with large-scale environments such as cities people combine direct experience and indirect sources such as maps. To ascertain which type of spatial knowledge is acquired by which source is difficult to evaluate. Using virtual reality enables the possibility to investigate whether knowledge is learned by direct experience or the use of a map differentially. Therefore, we designed a large virtual city, comprised of over 200 houses, and evaluated spatial knowledge acquisition after city exploration with an interactive map following one and three 30-min exploration sessions. We tested subjects' knowledge of the orientation of houses facing directions toward cardinal north, of orientations of houses facing directions relative to each other and pointing from one house to another. Our results revealed that increased familiarity after extended exploration with the map improved task accuracy. Further, it revealed task differences, caused mainly by a better accuracy in the relative orientation task than the pointing task. Time for cognitive reasoning improved overall task accuracy. Learning with our VR city map revealed an absence of distance effect, an alignment effect of tested house orientation toward map north and an angular difference effect between tested stimuli. Self-reported knowledge of cardinal directions learned in the real environment was positively correlated with task accuracy testing houses orientations toward cardinal north. Overall, our results suggest that participants learned spatial information that is directly available in the interactive map, while a spatial task that needed integration of learned knowledge stayed at lower accuracy levels.

Eye Tracking in Virtual Reality.

The intent of this paper is to provide an introduction into the bourgeoning field of eye tracking in Virtual Reality (VR). VR itself is an emerging technology on the consumer market, which will create many new opportunities in research. It offers a lab environment with high immersion and close alignment with reality. An experiment which is using VR takes place in a highly controlled environment and allows for a more in-depth amount of information to be gathered about the actions of a subject. Techniques for eye tracking were introduced more than a century ago and are now an established technique in psychological experiments, yet recent development makes it versatile and affordable. In combination, these two techniques allow unprecedented monitoring and control of human behavior in semi-realistic conditions. This paper will explore the methods and tools which can be applied in the implementation of experiments using eye tracking in VR following the example of one case study. Accompanying the technical descriptions, we present research that displays the effectiveness of the technology and show what kind of results can be obtained when using eye tracking in VR. It is meant to guide the reader through the process of bringing VR in combination with eye tracking into the lab and to inspire ideas for new experiments.