Adaptive, integrated sensor processing to compensate for drift and uncertainty: a stochastic ‘neural’ approach

Abstract

An adaptive stochastic classifier based on a simple, novel neural architecture -the Continuous Restricted Boltzmann Machine (CRBM) is demonstrated. Together with sensors and signal conditioning circuits, the classifier is capable of measuring and classifying (with high accuracy) the H + ion concentration, in the presence of both random noise and sensor drift. Training on-line, the stochastic classifier is able to overcome significant drift of real incomplete sensor data dynamically. As analogue hardware, this signal-level sensor fusion scheme is therefore suitable for real-time analysis in a miniaturised multisensor microsystem such as a Lab-in-a-Pill (LIAP).

An Implementation of a Spike-Response Model With Escape Noise Using an Avalanche Diode

“…Furthermore, noise-mediated effects, such as stochastic resonance [14], and noise-driven artificial neural networks, such as the Boltzmann machine [13], and its derivatives [47] need independent noise sources. If the promise of, for example, the continuous restricted Boltzmann machine as an embedded sensorfusion architecture [48] is to be fulfilled, multiple, integrated, compact, analog noise sources will be required. The SPAD-derived device described in this paper takes a fist step toward that hitherto-elusive goal.…”

Adaptive Microsensor Systems

“…This determination allows the model to adapt the classification rule depending on where a sample lies in feature space; such an approach may be used to adapt the classifier to the local environment of the instrument (e.g., temperature and humidity). More recently, Tang et al (126) proposed a stochastic neural network, the continuous restricted Boltzmann machine (CRBM), to perform binary classification with chemical sensor arrays in the 1 A classifier is said to be a weak learner if its performance is only slightly better than chance (e.g., a 51% success rate for a binary classification problem). Such learners are also typically unstable in the sense that small perturbations in the training set can lead to large changes in the learned parameters.…”