Embedded Classification of Local Field Potentials Recorded from Rat Barrel Cortex with Implanted Multi-Electrode Array

Abstract

This paper focuses on ultra-low power embedded classification of neural activities. The machine learning (ML) algorithm has been trained using evoked local field potentials (LFPs) recorded with an implanted 16×16 multi-electrode array (MEA) from the rat barrel cortex while stimulating the whisker. Experimental results demonstrate that ML can be successfully applied to noisy single-trial LFPs. We achieved up to 95.8% test accuracy in predicting the whisker deflection. The trained ML model is successfully implemented on a low-power embedded system with an average consumption of 2.6 mW.

Classification of Whisker Deflections From Evoked Responses in the Somatosensory Barrel Cortex With Spiking Neural Networks

“…This is in contrast to the standard approach where specifically engineered features are extracted from the raw signal. Interestingly, we found that random forests—a state-of-the art classifier—applied on features that were previously proposed (Temereanca and Simons, 2003 ; Mahmud et al, 2016 ; Wang et al, 2018 ) performed clearly worse than a LSM trained directly on MUA or LFP input. Although our statistical analysis showed that standard features—in particular those extracted from the LFP—do carry information about the stimulus, this indicates that SNNs can utilize information in the signal that is hidden when such features are considered.…”

“…In order to understand what components of LFPs and MUAs carry relevant information about whisker stimulation intensity, we first performed a statistical analysis on features of the signals that were selected based on our own experimental experience and on previous work (Wang et al, 2018 ).…”

“…The performance of LSTMs was considerably worse (LSTM, 68.9 ± 6.5%). Previous studies (Temereanca and Simons, 2003 ; Mahmud et al, 2016 ; Wang et al, 2018 ) relied on engineered features that were extracted from the LFP signal. We wondered whether such hand-crafted features could improve classification performance.…”

“…Then, the RPA was computed as the sum in voltage of the RP and the response's amplitude at the previously computed onset latency. Due to the slower dynamic characterizing the second part of the evoked LFP response, the PR , where present, was computed as the maximum positive value in voltage detected within the 100 ms following the main response peak, as described in Wang et al ( 2018 ).…”

HR-SAR-Net: A Deep Neural Network for Urban Scene Segmentation from High-Resolution SAR Data

“…The frequency of the transmitted electromagnetic signal defines the penetration depth into the soil, and the bandwidth tunes the geometric resolution. SAR is particularly interesting for a wide range of applications as it isas opposed to visual or multispectral imaging methodsresilient to weather conditions and cloud coverage, and it is independent of the lighting conditions (night, dusk/dawn) [7]. By a mere overflight of such a sensor either aboard a satellite, airplane, or drone, data can be collected to monitor crop growth, ocean wave height, floating icebergs, biomass estimation, snow monitoring, and maritime vessel tracking.…”