How good is good enough? The Approximate Computing Paradigm in Hardware, Software and Algorithms
Approximate computing is an emerging paradigm in computer science which spans the whole system stack from the hardware up to the software and the algorithms.
The goal of approximate computing is to exploit the error tolerance and robustness of certain applications in order to trade off precision against power, energy, storage, bandwidth and performance. This goal is achieved by both relaxing strict requirements on accuracy and precision, and by allowing a deviating behavior from exact (e.g. Boolean) specifications to some extent.
Besides the development of innovative new approximate hardware structures, research currently focuses on appropriate error metrics, as well as on new algorithms for approximate computing. In addition, the application of fault tolerance measures is investigated to enable a broadened spectrum of applications and to assure the quality of computational results.
This seminar approaches approximate computing at all levels - hardware, software and algorithm - and covers some of the most important challenges in the context of:
- Architecture and hardware design
- Technology and hardware implementation
- Approximate software techniques
- New algorithms for approximate computing
- Error metrics and characterization
- Fault tolerance for approximate computing
Schedule
Preliminary schedule (subject to changes):
Date | Room | Topic |
|---|---|---|
17.10.2016 | 0.363 | Introduction, topic overview and assignment (Slides) |
24.10.2016 | 0.363 | Session 0: Introduction to Approximate Computing |
05.12.2016 | 0.363 | Session 1: Metrics and Characterization in Approximate Computing |
12.12.2016 | 0.363 | Session 2: Programming and Software Techniques in Approximate Computing |
19.12.2016 | 0.363 | Session 3: Software Applications in Approximate Computing |
09.01.2017 | 0.363 | Session 4: Design and Synthesis / Neural Approximate Computing |
16.01.2017 | 0.363 | Session 5: Approximate and Precision-configurable Arithmetic Hardware |
23.01.2017 | 0.363 | Session 6: Memory in Approximate Computing |
30.01.2017 | 0.363 | –– |
06.02.2017 | 0.363 | –– |