Field of research
Many applications may perform inaccurate (or even erroneous) computations and still produce acceptable results. These applications show an inherent resiliency against inaccuracies since they:
- are designed to deal with noisy, imprecise, or incomplete input data,
- target human perception (we cannot identify small differences in an image or audio signals),
- cannot compute an exact or optimal result (e.g., a simulation which relies on models and heuristics, or accurate computations result in extremely long runtimes), or
- cannot compute the correct answer (e.g., for a recommender system, such as a search engine, a correct answer does not necessarily exist).
The design paradigm Approximate Computing exploits the resiliency of applications through the intentional design of incorrect software and hardware to trade off accuracy (or quality) against a target metric (e.g., runtime, energy consumption, or hardware area). However, it is vital that the approximate application meets the minimum quality constraints and ensures that the produced results are indeed acceptable. The application's user or designer usually defines the quality constraints. The images below show an example of outputs produced by a JPEG compression hardware and motivate the importance of applying approximations intelligently to produce acceptable results while achieving the best-possible improvements.
The image on the left-hand side shows the compressed output image generated with the original, precise system.
The center image and the image on the right show examples of approximate outcomes.
JPEG compression hardware subjected to approximations produced these two images. Some bits have been truncated from the computations for the hardware generating the center image, resulting in an outcome beyond acceptable quality. The approximations of the hardware creating the image on the right-hand side have been applied more intelligently, making it almost indistinguishable from the reference (left image). For both the image in the middle and the image on the right-hand side, power savings of around 53% over the original implementation have been achieved, highlighting the benefits of Approximate Computing and the advantages of the approximation technique that generated the image on the right.
