Current web-based visualizations are designed for single computers and cannot make use of additional devices on the client side, even if today’s users often have access to several, such as a tablet, a smartphone, and a smartwatch. We present a framework for ad-hoc computational clusters that leverage these local devices for visualization computations. Furthermore, we present an instantiating JavaScript toolkit called VisHive for constructing web-based visualization applications that can transparently connect multiple devices---called cells---into such ad-hoc clusters---called a hive---for local computation. Hives are formed either using a matchmaking service or through manual configuration. Cells are organized into a master-slave architecture, where the master provides the visual interface to the user and controls the slaves, and the slaves perform computation. VisHive is built entirely using current web technologies, runs in the native browser of each cell, and requires no specific software to be downloaded on the involved devices. We demonstrate VisHive using four distributed examples: a text analytics visualization, a database query for exploratory visualization, a

We present TimeFork, an interactive prediction technique to support users predicting the future of time-series data, such as in financial, scientific, or medical domains. TimeFork combines visual representations of multiple time series with prediction information generated by computational models. Using this method, analysts engage in a back-and-forth dialogue with the computational model by alternating between manually predicting future changes through interaction and letting the model automatically determine the most likely outcomes, to eventually come to a common prediction using the model. This computer-supported prediction approach allows for harnessing the user’s knowledge of factors influencing future behavior, as well as sophisticated computational models drawing on past performance. To validate the TimeFork technique, we conducted a user study in a stock market prediction game. We present evidence of improved performance for participants using TimeFork compared to fully manual or fully automatic predictions, and characterize qualitative usage patterns observed during the user study.