Modern workloads such as machine learning, multimedia processing, data mining show an intrinsic resilience to errors. Their characteristics such as redundancies in their input data and robust-to-noise algorithms or existence of multiple acceptable results allow them to produce outputs of acceptable quality, despite an approximation in some of their computations.

Approximate computing is an emerging design paradigm that leverages applications error resilience by relaxing traditional computation to a limited extent towards improving energy efficiency and performance. Techniques such as skipping non-critical computations at software level, or reducing circuit complexity and lowering the operation voltage at hardware level significantly improve performance in terms of execution time, area, and power/energy. In the era of lower power requirements and increasing computational complexity, they offer promising answers to both low power and high performance systems.

In this seminar we are interested in studying the applicability of Approximate Computing in all the abstraction layers, from operating systems and applications to architecture and arithmetic components, keeping a particular interest in the cross-layer interaction of approximations and also methodologies and tools for their implementation and analysis.