A Cognitive Stochastic Machine Based on Bayesian Inference: A Behavioral Analysis

Abstract

Bayesian models and stochastic computing form a promising paradigm for non-conventional, bio-inspired computation architectures. In particular, they are able to handle uncertainty and promise low power consumption. In this paper we study the application of such an architecture, the Sliced Bayesian Machine (SlicedBM) to a real-world problem, Sound Source Localization (SSL) for robots. We present an analysis of the quality of results and of computing time according to several parameters: sensor precision, result threshold, internal word length. Furthermore, we show that sensor data precision does not heavily influence the computation. On the opposite, the precision of the probability values plays an important role on result quality. This parameter also determines the circuit size. We also show that the higher the re-sampling threshold (RT), the better the distribution computed by the machine. Our results make it possible to choose optimal design parameters for a circuit along several trade-offs, and according to a given sensor fusion application.

Bayesian time-domain multiple sound source localization for a stochastic machine

“…A keypoint here is that our stochastic machine computes the posterior probability values (3) for all candidate values of s on the grid S, in parallel and in a very efficient/rapid manner, as explained in the upcoming sections. Indeed, the inference is basically done by multiplication of the different evidences, and our SM is precisely dedicated to perform matrix multiplications on probabilistic variables in a very efficient way, using AND-gates [7], [8].…”

“…Mono-source localization with a stochastic machine applied to signals pre-processed in the time-frequency (TF) domain was presented in [8] and deeply analyzed in [7]. However, one keypoint of stochastic machines is that they should avoid as much as possible signal pre-processing and focus directly on the Bayesian inference process.…”