Approximate Computation using Neuralized FPU

Neural networks can be used as function approximators to improve the energy efficiency, performance, and fault-tolerance of traditional computer architectures. To maximize these improvements the granularity of the function must be as large as possible. This work-inprogress abstract explores the lower limits of neural network function approximation by replacing individual floating point multiplications with multilayer perceptron neural networks. We show that this fine-grained approximation technique provides application dependent output accuracy for multiple applications in the PARSEC benchmark suite across varying network topologies.