Multiplicative recurrent neural network for fast and robust intracortical brain machine interface decoders
A brain machine interface (BMI) to control a device is provided. The BMI has a neural decoder, which is a neural to kinematic mapping function with neural signals as input to the neural decoder and kinematics to control the device as output of the neural decoder. The neural decoder is based on a continuous-time multiplicative recurrent neural network, which has been trained as a neural to kinematic mapping function. An advantage of the invention is the robustness of the decoder to perturbations in the neural data; its performance degrades less—or not at all in some circumstances—in comparison to the current state decoders. These perturbations make the current use of BMI in a clinical setting extremely challenging. This invention helps to ameliorate this problem. The robustness of the neural decoder does not come at the cost of some performance, in fact an improvement in performance is observed.
1. A brain-machine interface system, comprising:
at least one multi-electrode array implanted into a user's brain;
a prosthetic device; and
a neural signal decoder implemented using a computing device in communication with the at least one multi-electrode array and the prosthetic device, where the neural signal decoder comprises a recurrent neural network trained by:
obtaining a plurality of neural signals from a user using the at least one multi-electrode array, where:
the plurality of neural signals are obtained over the course of at least two days; and
the plurality of neural signals is modified by randomly adding and removing neural spikes such that the mean number of neural spikes is preserved;
instantiating the recurrent neural network, where the recurrent neural network is configured to map neural signals to kinematic functions of the prosthetic device; and;
training the recurrent neural network using the plurality of neural signals; and
where the neural signal decoder is configured to:
obtain a neural signal from the user using the at least one multi-electrode array describing an intended movement of the prosthetic device using the at least one multi-electrode array;
move the prosthetic device using the trained recurrent neural network to match the intended movement of the user.
2. The brain-machine interface system of claim 1 , wherein the prosthetic device is a robotic arm.
3. The brain-machine interface system of claim 1 , wherein the prosthetic device is a virtual cursor.
4. The brain-machine interface system of claim 1 , wherein the at least one multi-electrode array is implanted into the motor cortex of the user.
5. The brain-machine interface system of claim 1 , wherein the recurrent neural network is a multiplicative recurrent neural network.
6. The brain-machine interface system of claim 1 , wherein the recurrent neural network is retrained using a second plurality of neural signals.
7. A method for brain-machine interfacing, comprising:
obtaining a plurality of neural signals from a user using at least one multi-electrode array, where:
the plurality of neural signals are obtained over the course of at least two days; and
the plurality of neural signals is modified by randomly adding and removing neural spikes such that the mean number of neural spikes is preserved;
instantiating a recurrent neural network configured to map neural signals to kinematic functions of a prosthetic device;
training the recurrent neural network using the plurality of neural signals;
integrating the trained recurrent neural network into a brain machine interface, where the brain machine interface is in communication with the prosthetic device and the at least one multi-electrode array;
obtaining a neural signal from the user using the at least one multi-electrode array describing an intended movement of the prosthetic device using the at least one multi-electrode array; and
moving the prosthetic device using the trained recurrent neural network to match the intended movement of the user.
8. The method for brain-machine interfacing of claim 7 , wherein the prosthetic device is a robotic arm.
9. The method for brain-machine interfacing of claim 7 , wherein the prosthetic device is a virtual cursor.
10. The method for brain-machine interfacing of claim 7 , wherein the at least one multi-electrode array is implanted into the motor cortex of the user.
11. The method for brain-machine interfacing of claim 7 , wherein the recurrent neural network is a multiplicative recurrent neural network.
12. The method for brain-machine interfacing of claim 7 , further comprising:
obtaining a second plurality of neural signals; and
retraining the recurrent neural network using the second plurality of neural signals.
13. A method for brain-machine interfacing, comprising:
obtaining a first plurality of neural signals from a user using at least one multi-electrode array on a first day;
obtaining a second plurality of neural signals from the user using the at least one multi-electrode array on a second day;
generating a training data set comprising the first plurality of neural signals and the second plurality of neural signals, where the training data set is modified by randomly adding and removing neural spikes such that the mean number of neural spikes is preserved;
instantiating a recurrent neural network configured to map neural signals from the user to control functions of a prosthetic device;
training the recurrent neural network using the training data set;
integrating the trained recurrent neural network into a brain machine interface, where the brain machine interface is in communication with the prosthetic device and the at least one multi-electrode array;
obtaining a neural signal from the user using the at least one multi-electrode array describing an intended movement of the prosthetic device using the at least one multi-electrode array; and
moving the prosthetic device using the trained recurrent neural network to match the intended movement of the user.
14. The method of claim 13 , wherein the control functions are kinematic control functions.