Title : Evolving the keys to visual crowding
| Presenter | Erik van der Burg |
| Abstract | Peripheral vision can be severely impaired by nearby clutter. Decades of research using sparse displays have established that this phenomenon, known as visual crowding, follows Bouma’s rule: Interference occurs for target-distractor separations up to half the target’s eccentricity. Although considered a fundamental constraint on human vision, it is unclear whether Bouma’s rule holds in dense heterogeneous visual environments. Using a genetic algorithm we investigated crowding in densely cluttered displays. Participants were instructed to identify the orientation of a target line (6° eccentricity) among 284 distractor lines. Displays supporting highest accuracy were selected (“survival of the fittest”) and combined to create new displays. Performance improved over generations, predominantly driven by the emergence of horizontal flankers within 1° of the near-vertical target, but with no evidence of interference beyond this radius. We conclude that Bouma’s rule does not necessarily hold in densely cluttered displays. Instead, a nearest-neighbor segmentation rule provides a better account. |
Title : User-Driven Pattern Mining on knowledge graphs: an Archaeological Case Study
| Presenter | Xander Wilcke |
| Abstract | In recent years, there has been a growing interest from the Digital Humanities in knowledge graphs as data modelling paradigm. Already, many data sets have been published as such and are available in the Linked Open Data cloud. With it, the nature of these data has shifted from unstructured to structured. This presents new opportunities for data mining. In this work, we investigate to what extend data mining can contribute to the understanding of archaeological knowledge, expressed as knowledge graph, and which form would best meet the communities' needs. A case study was held which involved the user-driven mining of generalized association rules. Experiments have shown that the approach yielded mostly plausible patterns, some of which were seen as highly relevant by domain experts. |