Visualization-based Analytics: Presentations
Tapestry is a real-time distributed network visibility and analytics application that gathers data about endpoint-interactions from Domain Name System (DNS) servers, analyzes them, and computes an indicator called Network Complexity Index (NCI). It utilizes the inherently graphical structure of endpoint-interactions and provides a deeper level of visibility into networks by showing how endpoints are interacting with other endpoints, addressing many use cases in network security and network management.
Partitioning on a dimension in a dataset produces a small multiple display, an important technique for exploring multidimensional datasets. The problem of selecting a partitioning dimension to explain visual structure involves unguided search. We describe a set of goodness criteria for the collection of partitions in a small multiple. Then, we propose a method, based on non-parametric permutation tests, to evaluate the quality of various small multiples on real-world datasets.
We have developed segmentation-based methods to automatically cluster retinal images, and to measure the difference between our clustering and a gold standard created by a physician. The objective is to generate a clustering that closely approximates the gold standard. Our results show that our method produces clusterings that closely match the gold standard, and that after a certain point, finer segmentations fail to yield better approximations.