2. Related Work

While there has been a considerable body of research targeted at accelerating artificial neural networks (ANN) in general, very little work has been directed towards the architectural needs of perception processing and low power implementations of perception functions. Related areas of research can be classified broadly into optimization and characterization of perception workloads and various special purpose, parallel and low power coprocessor architectures. The following sections highlight representative work in each of these areas. In addition, general research in processor and reconfigurable logic not specifically targeted at perception yet capable of sustained high performance at low power budgets will also be discussed.


Binu Mathew