The notion of output quality is fundamental to approximate computing. It’s the basis of how we think about and evaluate research on approximation techniques. Because it’s so crucial to getting anything done, researchers tend to assume that quality is definable and measurable. But for many applications, quality is a fuzzy notion: it’s not clear at all how programmers can capture contextual, human reactions to output errors.
I recently gave a short talk about the problem of measuring quality in approximate applications. In keeping with my recent experiment with recording the talks I prepare, I made a five-minute video for your entertainment. Be sure to hit that thumbs-up button so I can go viral!