Mapping Inference Algorithms to DIMA

This chapter shows that diverse algorithms with significantly complex data-flow can also be mapped to DIMA. The mapping of a convolutional neural network (CNN) and a sparse distributed memory (SDM) to DIMA is demonstrated. Algorithmic opportunities such as the use of error-aware training in a DIMA-based CNN and the use of ensemble decision-making in SDM can be exploited to compensate for non-ideal analog computations in DIMA leading to even greater energy savings.