Ensemble methods for boosting the decoding performance of a hand grasping BMI

Abstract

Event Abstract Back to Event Ensemble methods for boosting the decoding performance of a hand grasping BMI Erk Subasi1*, Benjamin Townsend2 and Hans Scherberger2 1 Institute of Neuroinformatics, Computer Science, Switzerland 2 Institute of Neuroinformatics, Switzerland We are developing a brain machine interface (BMI) to decode grasp shape in macaque monkeys online. Neural activity is evaluated using chronically implanted electrodes in the anterior intraparietal cortex (AIP) and ventral premotor cortex (F5), areas known to be involved in the transformation of visual signals into hand grasping instructions. Macaque monkeys were trained in a delayed grasping task, where they first placed their hands at rest and fixated a red LED before a grasping handle was presented in one of 5 different orientations, and the color of an LED instructed the animals to grasp the handle either with a power or precision grip, respectively. After a short delay the fixation LED dimmed instructing the monkey to perform the required grasp. Correct trials were rewarded with a small amount of juice. After successful training 5 floating micro-electrode arrays (FMA; MicroProbe Inc) were implanted in AIP (2) and F5 (3) of 1 animal. Each array comprised 16 platinum-iridium electrodes (length 1.0-4.5mm, spacing 0.5 mm). This configuration was chosen to facilitate the recording of neuronal activity within cortical sulci instead of on the cortical surface. Neural signals were sampled using a Cerebus (Cyberkinetics Inc, Foxborough, MA) Neural Signal Processor (NSP) and streamed to a dedicated decoding PC via UDP. We previously reported our results using a standard maximum a posteriori (MAP) classifier where we obtained 94% performance for grasp type decoding (precision vs. power), following a mean accuracy of 91% for grasp type and 2 grasp orientations (target tilted to left or right) and finally a mean accuracy of 55% where we decode full 10 conditions (grasp type and 5 orientations). MAP decoding is one of the standard methods in BMI literature which is shown yielding robust decoding results in similar experiments before. But we make 3 common strong assumptions regarding the underlying probability distributions in MAP framework. First, we assume a Poisson probability distribution for firing rates of each individual unit. Second, in order to ease the joint probability distribution computation, it is assumed that neurons are firing in an independent fashion. And third, it is assumed that these units have stationary firing probability distributions. It is important to note that a marginal increase in the decoding performance of a neural-prosthetic device may provide substantial improvement in perceived quality of the system by the user. Here, we introduce a new approach for decoding discrete grasping postures which does not need to rely on any of these assumptions. Ensemble methods have gained popularity among the machine learning community in the last decade, both due to their strong theoretical base and practical outcomes. They mainly try to optimally utilize many weak learners on a training set for achieving better performance compared to a single strong classifier while reducing the variance of the classifier and in some cases also providing a better bias. Here, the comparative decoding analysis will be presented with adaptive bagging and boosting techniques. Conference: Neuroinformatics 2009, Pilsen, Czechia, 6 Sep - 8 Sep, 2009. Presentation Type: Poster Presentation Topic: Brain machine interface Citation: Subasi E, Townsend B and Scherberger H (2019). Ensemble methods for boosting the decoding performance of a hand grasping BMI. Front. Neuroinform. Conference Abstract: Neuroinformatics 2009. doi: 10.3389/conf.neuro.11.2009.08.056 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 22 May 2009; Published Online: 09 May 2019. * Correspondence: Erk Subasi, Institute of Neuroinformatics, Computer Science, Zurich, Switzerland, erk@ini.phys.ethz.ch Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Erk Subasi Benjamin Townsend Hans Scherberger Google Erk Subasi Benjamin Townsend Hans Scherberger Google Scholar Erk Subasi Benjamin Townsend Hans Scherberger PubMed Erk Subasi Benjamin Townsend Hans Scherberger Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.