High level neuromorphic network description apparatus and methods

Patent metadata

Field	Value
Publication number	US-10210452-B2
Application number	US-201213385933-A
Country	US
Kind code	B2
Filing date	Mar 15, 2012
Priority date	Sep 21, 2011
Publication date	Feb 19, 2019
Grant date	Feb 19, 2019

Abstract

Apparatus and methods for high-level neuromorphic network description (HLND) framework that may be configured to enable users to define neuromorphic network architectures using a unified and unambiguous representation that is both human-readable and machine-interpretable. The framework may be used to define nodes types, node-to-node connection types, instantiate node instances for different node types, and to generate instances of connection types between these nodes. To facilitate framework usage, the HLND format may provide the flexibility required by computational neuroscientists and, at the same time, provides a user-friendly interface for users with limited experience in modeling neurons. The HLND kernel may comprise an interface to Elementary Network Description (END) that is optimized for efficient representation of neuronal systems in hardware-independent manner and enables seamless translation of HLND model description into hardware instructions for execution by various processing modules.

First claim

What is claimed:

1. A method of implementing a neural network using an instruction set, the method comprising:
generating a graphical representation of a plurality of nodes of the neural network via a first panel of a graphical user interface (GUI);
converting the graphical representation to a high-level representation of the plurality of nodes to be displayed concurrently in a second panel of the GUI with the graphical representation, the high-level representation comprising a plurality of instructions of the instruction set, the plurality of instructions defining an architecture of the neural network including a node definition specifying an internal implementation of a node of the neural network including dynamics of a node type, the instruction set comprising a first instruction corresponding to at least one node of the plurality of nodes, the first instruction comprising a syntax based on natural English language structure and grammar to instantiate the at least one node within the neural network;
encoding the high-level representation into a low-level hardware independent format; and
operating the neural network based on the low-level hardware independent format.

2. The method of claim 1 , wherein the first instruction comprises a keyword selected from a group consisting of CREATE, MAKE, and PUT.

3. The method of claim 2 , wherein the keyword comprises only one keyword.

4. The method of claim 1 , wherein:
the at least one node is characterized by a spatial coordinate; and
the first instruction assigns a desired location within the neural network to the spatial coordinate, the assignment being performed as a part of the instantiating of the at least one node.

5. The method of claim 4 , wherein the first instruction comprises a keyword selected from a group consisting of ON, IN, and AT.

6. The method of claim 5 , wherein the keyword comprises only one keyword.

7. The method of claim 1 , wherein:
the instantiating of the at least one node comprises generating a first node and a second node; and
the instruction set comprises a second instruction to generate a connection between the first node and the second node.

8. The method of claim 7 , wherein the connection comprises a synapse.

9. The method of claim 7 , wherein the connection comprises a junction.

10. The method of claim 7 , wherein:
instantiating the at least one node further comprises generating a plurality of nodes, the plurality of nodes comprising the first node and the second node;
each node of the plurality of nodes comprising at least one tag to facilitate operation of the neural network; and
the instruction set includes a third instruction to select a subset of the plurality of nodes based on the at least one tag.

11. The method of claim 7 , wherein the second instruction comprises a keyword selected from a group consisting of CONNECT, LINK, and PROJECT.

12. The method of claim 7 , wherein the second instruction comprises a keyword selected from a group consisting of WITH, BY, and USING, the keyword adapted to specify a type of the connection.

13. The method of claim 1 , wherein:
the neural network comprises a plurality of elements, each element of the plurality of elements having a tag associated therewith; and
the first instruction identifies a subset of the plurality of elements, based at least in part on the tag.

14. The method of claim 13 , wherein the compiling is based on a database comprising a plurality of tags.

15. The method of claim 13 , wherein the instruction set comprises a second instruction to effect assignment of a new tag to the subset.

16. The method of claim 15 , wherein the second instruction comprises a keyword selected from a list consisting of TAG, ASSIGN, and MARK.

17. The method of claim 1 , wherein the instruction set further comprises a Boolean expression consistent with a structured English representation; and
the structured English representation enables machine execution of the Boolean expression by at least in part an implicit assignment of logical AND operation between any two Boolean variables of the Boolean expression, where the two Boolean variables are separated by a separator keyword thereby.

18. The method of claim 17 , wherein the separator keyword comprises a whitespace.

19. The method of claim 1 , wherein the node definition also defines connectivity of the node.

20. The method of claim 1 , further comprising:
compiling the low-level hardware independent format into machine executable instructions for execution on a specific hardware platform; and
executing the hardware dependent machine executable instructions to operate the neural network on the specific hardware platform.

21. A non-transitory computer readable medium having encoded thereon program code for implementing a neural network using an instruction set, the program code being executed by a processor and comprising:
program code to generate a graphical representation of a plurality of nodes of the neural network via a first panel of a graphical user interface (GUI);
program code to convert the graphical representation to a high-level representation of the plurality of nodes to be displayed concurrently in a second panel of the GUI with the graphical representation, the high-level representation comprising a first plurality of instructions of the instruction set, the first plurality of instructions defining an architecture of the neural network including a node definition, specifying an internal implementation of a node of the neural network including dynamics of a node type, the instruction set comprising a first instruction corresponding to at least one node of the plurality of nodes, the first instruction comprising a syntax based on natural English language structure and grammar to instantiate the at least one node within the neural network;
program code to encode the high-level representation into a low-level hardware independent format; and
program code to operate the neural network based on the low-level hardware independent format.

22. The non-transitory computer readable medium of claim 21 , further comprising program code:
to compile the low-level hardware independent format into machine executable instructions for execution on a specific hardware platform, and
to execute the machine executable instructions to operate the neural network on the specific hardware platform.

23. The non-transitory computer readable medium of claim 22 , wherein the low-level hardware independent format comprises Elementary Network Description (END) format.

24. The non-transitory computer readable medium of claim 23 , wherein the hardware dependent machine executable instructions are central processing unit (CPU) instructions, graphics processing unit (GPU) instructions, or field programmable gate array (FPGA) instructions.

25. The non-transitory computer readable medium of claim 22 , wherein the low-level hardware independent format comprises a plurality of functions for execution by a processing module.

26. The non-transitory computer readable medium of claim 22 , wherein the high-level representation comprises a second plurality of instructions received via a user speaking the second plurality of instructions.

27. The non-transitory computer readable medium of claim 26 , wherein speaking the second plurality of instructions is effected one instruction at a time.

28. The

Assignees

• Szatmary Botond
• Izhikevich Eugene M
• Petre Csaba
• Nageswaran Jayram Moorkanikara
• Piekniewski Filip
• Qualcomm Inc

Inventors

• Szatmary Botond
• Izhikevich Eugene M
• Petre Csaba
• Nageswaran Jayram Moorkanikara
• Piekniewski Filip

Classifications

• G06N3/105 Primary

  Shells for specifying net layout · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06N3/042

  Knowledge-based neural networks; Logical representations of neural networks · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title