Evidence of a mechanism of neural adaptation in the closed loop control of directions

Using cortical neurons of animals to control external devices allows experimenters a unique opportunity to study the capability of the brain to utilize a new actuator. The purpose of this paper is to investigate the ability of unrestrained rats to control a directional task using motor cortical signals.

In freely moving rats, signals recorded from the motor cortex of the brain enabled the use of a closed loop brain machine interface (BMI) system to replace paddle pressing for a directional task. In this system, ring rates from several (two to ten) motor cortical neurons at several consecutive time points were used as input to a support vector machine (SVM) classifier. The decision‐function value obtained from the SVM was then used to determine which relay should be activated to produce paddle‐pressing signals in the task. All five animals tested were able to use this interface immediately and significant changes in neural activity arise in a single, 45‐min, experimental session. Neural data from three of the subjects were examined for changes from the calibration phase (data used to build the SVM model) to the late cortically controlled phase.

Detailed analysis shows that neural activity changes significantly from the calibration phase to the cortically controlled phase, furthermore, the decision‐function values arising from these neural signals change to support better performance. By examining which neurons and times are selected by the SVM to have significant impact on the decision‐function value as well as which of these elements change significantly, a mechanism of adaptation begins to emerge in which the SVM properly assigns high importance to dimensions that easily predict the desired output, however, under closed‐loop control, the animal selects a small number of neurons (at most or all times) and chooses to make the firing rates more distinguishable. Video taken of one of the subjects further suggests the nature of the behavioral correlates of these changes.

In the design of practical BMI devices for human patients, one effective strategy might involve creating mappings from multi‐neuron ensembles using state‐of‐the‐art machine learning techniques, but expect that the patients who use the devices will adapt to the devices using single neuron modulation changes.

In the design of practical BMI devices for human patients, one effective strategy might involve creating mappings from multi‐neuron ensembles using state‐of‐the‐art machine learning techniques, but expect that the patients who use the devices will adapt to the devices using single neuron modulation changes. The proposed decentralization approach is interesting for the design of optimization algorithms that can run on computing systems that use principles of self‐organization and have no central control.