Approximate Systems

Motivation

Resource efficiency is becoming an increasingly important challenge, especially due to the pervasiveness of computing systems and the diminishing returns from performance improvements of process technology scaling. At the same time, many important applications have nondeterministic specifications or are robust to noise in their execution. They thus do not necessarily require fully reliable computing systems and their resource consumption can be reduced by introducing or exposing approximations. While trading correctness for efficiency has been part of computing systems since the early days, it has seen renewed interest in the past decade. Different techniques have been since developed for applying and controlling approximations and the errors they introduce at different levels of the compute stack. Unfortunately, most of these techniques have been applied in isolation, making simplified assumptions about the other levels. It is thus unclear how all the different techniques interact, combine and complement or negate each other to provide end-to-end application benefits. Future work is needed to investigate error-efficient or approximate computing approaches systematically and holistically across all layers of the compute stack and for complete end-to-end system design examples. This Dagstuhl Seminar aims to bring together researchers from the different domains working on approximate computing, algorithms, programming languages, compilers, architecture and circuits, in order to explore open challenges and opportunities and to define cross-area research directions relating to an end-to-end application of approximate computing principles across the compute stack. The seminar will be structured around four application areas and application examples in embedded sensing and computation, robotics and control, scientific computing, and machine learning, which will serve as the focal points of discussions.