Traditional computing imposes an “exactness” that is becoming both
energy expensive and unnecessary, especially so in the era of big data.
The relaxation of accuracy, and more importantly the relaxation of the
requirement of deterministic execution provide an opportunity to explore
new energy-efficient execution models to process this vast data. Recently,
the approximate computing paradigm has continued to gain importance
as a vehicle to reason about changes across the stack from algorithms to
devices to produce results that are acceptable albeit different from exact
computation.