Cost-performance tradeoffs in unreliable computation architectures

Abstract

We investigate fusing several unreliable computational units that perform the same task. We model an unreliable computational outcome as an additive perturbation to its error-free result in terms of its fidelity and cost. We analyze performance of repetition-based strategies that distribute cost across several unreliable units and fuse their outcomes. When the cost is a convex function of fidelity, the optimal repetition-based strategy in terms of incurred cost while achieving a target mean-square error (MSE) performance may fuse several computational units. For concave and linear costs, a single more reliable unit incurs lower cost compared to fusion of several lower cost and less reliable units while achieving the same MSE performance. We show how our results give insight into problems from theoretical neuroscience, circuits, and crowdsourcing.

Cost-Reliability Tradeoffs in Fusing Unreliable Computational Units

“…Note that some of the results in this paper appeared in an earlier version [17]. In particular, we originally presented the noisy computation model and the cost-fidelity formulation in [17] with some early analyses on different cost-fidelity functions.…”

“…Note that some of the results in this paper appeared in an earlier version [17]. In particular, we originally presented the noisy computation model and the cost-fidelity formulation in [17] with some early analyses on different cost-fidelity functions. In this paper, we extend the analyses in [17] further and provide a more detailed and rigorous treatment for different scenarios.…”

“…In particular, we originally presented the noisy computation model and the cost-fidelity formulation in [17] with some early analyses on different cost-fidelity functions. In this paper, we extend the analyses in [17] further and provide a more detailed and rigorous treatment for different scenarios. In addition, we present how the cost-fidelity formulation and the theoretical results can be applied to problems from neuroscience and circuits.…”

Boosting Classifiers with Noisy Inference

“…Our approach provides a general framework to allocate a limited resource for boosting classifiers and can also be applied to settings of noisy computations. For example, the quality of computations on noisy hardware can be changed by controlling supply voltage [8], replicating computations [12], [13], and implementing granular bit precisions [10]. Based on the proposed framework, we can optimize these system resources in a principled manner.…”