I’ve reading further afield from my “academic home turf” (numerical graph computations for large datasets). The present excursion is to probabilistic databases. I’ve never understood their use! It always seemed like a bunch of complexity for little realistic benefit. I’m still not convinced, but I’ll keep reading — that field has many elegant abstractions.
In the course of my reading, I came across a clever paper:
Exploiting Correlated Attributes in Acquisitional Query Processing by Amol Deshpande, Carlos Guestrin, Wei Hong, and Samuel Madden.
Suppose we query a set of low-power sensors to gather data about a set of circumstances. The example from the paper is that we want to collect data when it’s dark outside (< 100 lux) but still warm (> 20C). If taking a measurement of either light or temperature costs 1 unit of power (i.e. 2 units to check both), then its silly to check temperature during “daylight” hours (6am to 6pm) until we’ve checked if it’s dark enough outside. Likewise, at night, it’s silly to check light until we’ve checked the temperature. The goal is to construct simple rules to optimize “query” execution in these acquisitional environments. I didn’t bother reading the rest of the paper, but that example was just so clever!