Limits...
A Generalizable Brain-Computer Interface (BCI) Using Machine Learning for Feature Discovery.

Nurse ES, Karoly PJ, Grayden DB, Freestone DR - PLoS ONE (2015)

Bottom Line: The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge.Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure.Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.

View Article: PubMed Central - PubMed

Affiliation: NeuroEngineering Laboratory, Department of Electrical & Electronic Engineering, The University of Melbourne, Melbourne, VIC, Australia, 3010; Centre for Neural Engineering, The University of Melbourne, Melbourne, VIC, Australia, 3010.

ABSTRACT
This work describes a generalized method for classifying motor-related neural signals for a brain-computer interface (BCI), based on a stochastic machine learning method. The method differs from the various feature extraction and selection techniques employed in many other BCI systems. The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge. Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure. The results showed that the average performance of the new algorithm outperformed other published methods using the Berlin BCI IV (2008) competition dataset and was comparable to the best results in the Berlin BCI II (2002-3) competition dataset. The new method was also applied to electroencephalography (EEG) data recorded from five subjects undertaking a hand squeeze task and demonstrated high levels of accuracy with a mean classification accuracy of 78.9% after five-fold cross-validation. Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.

No MeSH data available.


Related in: MedlinePlus

Classification performance of an artificial neural network (ANN) using backpropagation (BP) and simulated annealing augmented backpropagation (SA).Two randomly generated one-, two- or three-layer ANNs were created. Both ANNs had the same number of hidden layers and neurons. Each ANN was then trained using either BP or SA and tested on the same unseen test set. This process was repeated 50 times. The bar represents the mean Cohen’s kappa score for each group and the error bars are the standard deviation of the 50 kappa scores. The outcome for chance performance is shown as a red dashed line (kappa score of zero). The asterisks indicate the confidence level with which to reject the hypothesis that the two bars are from the same distribution using a two-tailed t-test (** for p < 0.01). For exact p-values see S2 Appendex. Datasets are from the BCI II competition, dataset 4 and the BCI IV competition, dataset 3. The results for the two- and three-class hand-squeeze datasets are taken as the average values across five participants.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4482677&req=5

pone.0131328.g002: Classification performance of an artificial neural network (ANN) using backpropagation (BP) and simulated annealing augmented backpropagation (SA).Two randomly generated one-, two- or three-layer ANNs were created. Both ANNs had the same number of hidden layers and neurons. Each ANN was then trained using either BP or SA and tested on the same unseen test set. This process was repeated 50 times. The bar represents the mean Cohen’s kappa score for each group and the error bars are the standard deviation of the 50 kappa scores. The outcome for chance performance is shown as a red dashed line (kappa score of zero). The asterisks indicate the confidence level with which to reject the hypothesis that the two bars are from the same distribution using a two-tailed t-test (** for p < 0.01). For exact p-values see S2 Appendex. Datasets are from the BCI II competition, dataset 4 and the BCI IV competition, dataset 3. The results for the two- and three-class hand-squeeze datasets are taken as the average values across five participants.

Mentions: In order to verify the benefits of using SA in the classification algorithm, the performance of a network trained using scaled conjugate backpropagation with and without SA was assessed. A randomly generated one-, two- or three-layer network was trained on the data using both backpropagation and SA augmented backpropagation, and then its Cohen’s kappa score was evaluated on an unseen test set. This process was repeated for 50 realizations, using 10 iterations of SA. A two-tail t-test was used to determine whether the two methods could be considered as significantly different. Fig 2 shows the results of this comparison for both competition datasets and the two- and three-class hand squeeze task. It can be seen from Fig 2 that the standard deviation of the Cohen’s kappa score decreased with the use of SA-modified backpropagation and the mean performance significantly increased, for every dataset. However, the improvement appears more pronounced for datasets that contained less data (the BCI competitions).


A Generalizable Brain-Computer Interface (BCI) Using Machine Learning for Feature Discovery.

Nurse ES, Karoly PJ, Grayden DB, Freestone DR - PLoS ONE (2015)

Classification performance of an artificial neural network (ANN) using backpropagation (BP) and simulated annealing augmented backpropagation (SA).Two randomly generated one-, two- or three-layer ANNs were created. Both ANNs had the same number of hidden layers and neurons. Each ANN was then trained using either BP or SA and tested on the same unseen test set. This process was repeated 50 times. The bar represents the mean Cohen’s kappa score for each group and the error bars are the standard deviation of the 50 kappa scores. The outcome for chance performance is shown as a red dashed line (kappa score of zero). The asterisks indicate the confidence level with which to reject the hypothesis that the two bars are from the same distribution using a two-tailed t-test (** for p < 0.01). For exact p-values see S2 Appendex. Datasets are from the BCI II competition, dataset 4 and the BCI IV competition, dataset 3. The results for the two- and three-class hand-squeeze datasets are taken as the average values across five participants.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4482677&req=5

pone.0131328.g002: Classification performance of an artificial neural network (ANN) using backpropagation (BP) and simulated annealing augmented backpropagation (SA).Two randomly generated one-, two- or three-layer ANNs were created. Both ANNs had the same number of hidden layers and neurons. Each ANN was then trained using either BP or SA and tested on the same unseen test set. This process was repeated 50 times. The bar represents the mean Cohen’s kappa score for each group and the error bars are the standard deviation of the 50 kappa scores. The outcome for chance performance is shown as a red dashed line (kappa score of zero). The asterisks indicate the confidence level with which to reject the hypothesis that the two bars are from the same distribution using a two-tailed t-test (** for p < 0.01). For exact p-values see S2 Appendex. Datasets are from the BCI II competition, dataset 4 and the BCI IV competition, dataset 3. The results for the two- and three-class hand-squeeze datasets are taken as the average values across five participants.
Mentions: In order to verify the benefits of using SA in the classification algorithm, the performance of a network trained using scaled conjugate backpropagation with and without SA was assessed. A randomly generated one-, two- or three-layer network was trained on the data using both backpropagation and SA augmented backpropagation, and then its Cohen’s kappa score was evaluated on an unseen test set. This process was repeated for 50 realizations, using 10 iterations of SA. A two-tail t-test was used to determine whether the two methods could be considered as significantly different. Fig 2 shows the results of this comparison for both competition datasets and the two- and three-class hand squeeze task. It can be seen from Fig 2 that the standard deviation of the Cohen’s kappa score decreased with the use of SA-modified backpropagation and the mean performance significantly increased, for every dataset. However, the improvement appears more pronounced for datasets that contained less data (the BCI competitions).

Bottom Line: The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge.Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure.Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.

View Article: PubMed Central - PubMed

Affiliation: NeuroEngineering Laboratory, Department of Electrical & Electronic Engineering, The University of Melbourne, Melbourne, VIC, Australia, 3010; Centre for Neural Engineering, The University of Melbourne, Melbourne, VIC, Australia, 3010.

ABSTRACT
This work describes a generalized method for classifying motor-related neural signals for a brain-computer interface (BCI), based on a stochastic machine learning method. The method differs from the various feature extraction and selection techniques employed in many other BCI systems. The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge. Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure. The results showed that the average performance of the new algorithm outperformed other published methods using the Berlin BCI IV (2008) competition dataset and was comparable to the best results in the Berlin BCI II (2002-3) competition dataset. The new method was also applied to electroencephalography (EEG) data recorded from five subjects undertaking a hand squeeze task and demonstrated high levels of accuracy with a mean classification accuracy of 78.9% after five-fold cross-validation. Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.

No MeSH data available.


Related in: MedlinePlus