Biomedical subjects

Carl Gold

Publications and source records attributed to Carl Gold.

2 recordsLinked to original sources

PubMed2006

On the origin of the extracellular action potential waveform: A modeling study.

Although extracellular unit recording is typically used for the detection of spike occurrences, it also has the theoretical ability to report about what are typically considered intracellular features of the action potential. We address this theoretical ability by developing a model system that captures features of experimentally recorded simultaneous intracellular and extracellular recordings of CA1 pyramidal neurons. We use the line source approximation method of Holt and Koch to model the extracellular action potential (EAP) voltage resulting from the spiking activity of individual neurons. We compare the simultaneous intracellular and extracellular recordings of CA1 pyramidal neurons recorded in vivo with model predictions for the same cells reconstructed and simulated with compartmental models. The model accurately reproduces both the waveform and the amplitude of the EAPs, although it was difficult to achieve simultaneous good matches on both the intracellular and extracellular waveforms. This suggests that accounting for the EAP waveform provides a considerable constraint on the overall model. The developed model explains how and why the waveform varies with electrode position relative to the recorded cell. Interestingly, each cell's dendritic morphology had very little impact on the EAP waveform. The model also demonstrates that the varied composition of ionic currents in different cells is reflected in the features of the EAP.

Action Potentials↗

PubMedDate not supplied

Bayesian approach to feature selection and parameter tuning for support vector machine classifiers.

A Bayesian point of view of SVM classifiers allows the definition of a quantity analogous to the evidence in probabilistic models. By maximizing this one can systematically tune hyperparameters and, via automatic relevance determination (ARD), select relevant input features. Evidence gradients are expressed as averages over the associated posterior and can be approximated using Hybrid Monte Carlo (HMC) sampling. We describe how a Nyström approximation of the Gram matrix can be used to speed up sampling times significantly while maintaining almost unchanged classification accuracy. In experiments on classification problems with a significant number of irrelevant features this approach to ARD can give a significant improvement in classification performance over more traditional, non-ARD, SVM systems. The final tuned hyperparameter values provide a useful criterion for pruning irrelevant features, and we define a measure of relevance with which to determine systematically how many features should be removed. This use of ARD for hard feature selection can improve classification accuracy in non-ARD SVMs. In the majority of cases, however, we find that in data sets constructed by human domain experts the performance of non-ARD SVMs is largely insensitive to the presence of some less relevant features. Eliminating such features via ARD then does not improve classification accuracy, but leads to impressive reductions in the number of features required, by up to 75%.

Algorithms↗