NEURAL NETWORK LEARNING AND COLLABORATION APPARATUS AND METHODS

• US 20140089232A1
• Filed: 03/14/2013
• Published: 03/27/2014
• Est. Priority Date: 06/01/2012
• Status: Active Grant

• Current Assignee: Brain Corporation
• Original Assignee: Brain Corporation
• Inventors: Buibas, Marius, Izhikevich, Eugene M., Szatmary, Botond, Polonichko, Vadim
• Granted Patent: US 9,208,432 B2
• Time in Patent Office: Days
• US Class Current: 706/11
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/10 Interfaces, programming languages or software development kits, e.g. for simulating neural networks

Abstract

Apparatus and methods for learning and training in neural network-based devices. In one implementation, the devices each comprise multiple spiking neurons, configured to process sensory input. In one approach, alternate heterosynaptic plasticity mechanisms are used to enhance learning and field diversity within the devices. The selection of alternate plasticity rules is based on recent post-synaptic activity of neighboring neurons. Apparatus and methods for simplifying training of the devices are also disclosed, including a computer-based application. A data representation of the neural network may be imaged and transferred to another computational environment, effectively copying the brain. Techniques and architectures for achieve this training, storing, and distributing these data representations are also disclosed.

Claims

1. A method of behavioral programming in an artificial neural network, the method comprising:

generating a data link to at least one device configured to run the artificial neural network;

receiving one or more data elements indicating a current status associated with the artificial neural network;

causing display of information related to at least a portion of the one or more data elements;

receiving user input from a user interface;

generating one or more feedback elements based at least in part on the user input; and

transmitting the one or more feedback elements to the artificial neural network via the data link.

2. The method of claim 1, wherein the at least one device configured to run the artificial neural network comprises a robot comprising one or more functional modules.

3. The method of claim 2, wherein the one or more functional modules comprise at least one sensory device.

4. The method of claim 3, wherein:

the artificial neural network is configured to generate one or more discrete output signals in response to input from the at least one sensory device; and

the discrete output signals cause an action at least one of the one or more functional modules.

5. The method of claim 4, wherein:

the displayed information comprises an one or more visual representations of the input from the sensor; and

at least one indicator configured to indicate one or more neurons involved in the generation of the one or more discrete output signals.

6. The method of claim 1, wherein the display and the user interface are disposed on a mobile wireless-enabled device.

7. The method of claim 6, wherein the generated data link is effected via at least a personal area network.

8. The method of claim 6, wherein the generated data link comprises a data session over at least a Wi-Fi compliant connection.

9. The method of claim 1, wherein the one or more feedback elements comprise negative feedback designed to suppress an output generated by one or more neurons of the artificial neural network.

10. The method of claim 1, wherein the one or more feedback elements comprise positive feedback to potentiate one or more neurons of the artificial neural network.

11. The method of claim 1, further comprising downloading an image of the artificial neural network to the at least one device from a cloud-based server.

12. The method of claim 11, further comprising causing recommendation of the image via at least one application running on the cloud-based server, the recommendation being based at least in part on a task designated by the user.

13. A non-transitory computer readable medium configured to store at least one computer program thereon, the computer program comprising a plurality of instructions configured to when executed:

establish a data connection to a synaptic device;

receive status information from the synaptic device;

issue a command to the synaptic device, the synaptic device executing an action based at least in part on the command;

receive feedback input from a user; and

forward the feedback input to the synaptic device via the data connection;

wherein the act of forwarding feedback input causes the synaptic device to alter a behavioral trait.

14. The computer readable medium of claim 13, wherein the instructions are further configured to, when executed, present the user with one or more menus from which to select the feedback input.

15. The computer readable medium of claim 13, wherein the instructions are further configured to, when executed, display one or more of audio and visual representations of the status information.

16. The computer readable medium of claim 13, wherein the alteration of the behavioral trait comprises an adjustment of a configuration of an artificial neural network disposed at least in part on the synaptic device.

17. A training device configured to manage activity in at least one spiking neural network, the training device comprising:

at least one network interface configured to;

establish an operative link to the spiking neural network;

receive one or more activity indicators associated with the at least one spiking neural network; and

transmit feedback input to the at least one spiking neural network;

a user interface configured to;

based at least in part on the one or more activity indicators, display one or more human perceptible signals; and

provide a user with at least one menu from which to select training options; and

logic in operative communication with the user interface and network interface, the logic configured to process a selected training option to generate the feedback input.

18. The training device of claim 17, wherein the training device comprises a mobile wireless device coupled to the at least one spiking neural network via the network interface.

19. The training device of claim 18, wherein the operative link comprises one or more of a wireless local area network (WLAN) link and a personal area network (PAN) link.

20. The training device of claim 17, wherein the device is further configured to obtain an image of the at least one spiking neural network.

21. The training device of claim 17, wherein the feedback input alters one or more states of a plurality of neurons in the at least one spiking neural network.

22. A method of managing activity within a spiking neural network, the method comprising:

establishing a data link to a user interface device;

sending one or more status updates related to a plurality of activity states neurons in the spiking neural network;

receiving, via the link, one or more feedback indicators, the one or more feedback indicators based at least on part on a selected training option from the user interface device; and

based on at least one rule, associating the feedback with a subset of the neurons in the spiking neural network.

23. The method of claim 22, wherein associating the feedback comprises potentiating the subset of neurons.

24. The method of claim 22, wherein associating the feedback comprises depressing the subset of neurons.

25. The method of claim 22, wherein spiking neural network is at least partially disposed on a computerized neuromorphic apparatus in operative communication with the user interface device.

26. A computerized neuromorphic apparatus comprising:

one or more functional modules;

a network interface configured to establish a link to a training device; and

logic configured to;

send, to the training device, status indicators related to a neural network disposed at least in part on the computerized neuromorphic apparatus;

receive feedback based on a selected one or more of a plurality of available management options; and

alter a state of one or more of a neuron and a connection in accordance with the feedback and at least one timing rule.

27. The method of claim 26, wherein:

a magnitude and a direction of the alteration are based on at least the feedback; and

a target of the alteration is selected based on the at least one timing rule.

28. The method of claim 26, wherein:

the status indicators comprises an identified action and an identifier for one or more contributory neurons; and

the plurality of available management options comprise a positive feedback option and a negative feedback option.