Round-trip engineering apparatus and methods for neural networks

• US 9,117,176 B2
• Filed: 03/15/2012
• Issued: 08/25/2015
• Est. Priority Date: 09/21/2011
• Status: Expired due to Fees

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm Technologies Incorporated (Qualcomm, Inc.)
• Inventors: Piekniewski, Filip, Szatmary, Botond, Izhikevich, Eugene M., Petre, Csaba, Nageswaran, Jayram Moorkanikara
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): OLUDE AFOLABI, OLATOYOSI
• Application Number:
  US13/385,937
• Publication Number:
  US 20130218821A1
• Time in Patent Office: 1,258 Days
• Field of Search: 706/25, 706/45
• US Class Current: 1/1
• CPC Class Codes: G06F 8/355 Round-trip engineering
• 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 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.

Claims

1. Computer readable apparatus comprising a non-transitory storage medium, said storage medium comprising at least one computer program with a plurality of instructions, said at least one program being capable of implementing round-trip engineering method in a neural network by at least:

enabling receiving user inputs via a graphical user interface;

based at least in part on first user input indicative of a plurality of nodes within said network, automatically generating first single network description instruction;

based at least in part on second user input indicative of a tag, automatically generating second single network description instruction; and

based at least in part on third user input indicative of node connections, automatically generating third single network description instruction; and

wherein;

said first, second and third single instructions, when executed, are configured to;

(i) generate said plurality of nodes;

(ii) select a portion of said plurality of nodes based on said tag, and (iii) generate a plurality of connections associated, at least in part, with said portion of said plurality of nodes, substantially without relying on subsequent user input.

2. The apparatus of claim 1, wherein said plurality of connections is distributed within the network in accordance with a probability distribution comprising a spatial parameter.

3. The apparatus of claim 1, wherein said method further comprises assigning one other tag to said plurality of connections, said one other tag being distinct from said tag.

4. The apparatus of claim 1, wherein said each said first, second and third single instruction is configured consistent with English language structure and grammar.

5. Computer readable apparatus comprising a non-transitory storage medium, said storage medium comprising a plurality of computer instructions being capable of implementing round-trip engineering (RTE) method in a computerized neural network, the method comprising:

receiving first user input indicative of generation of a plurality of nodes within said network; and

based at least in part on said input, generating a single network description instruction, which, when executed, causes generation of said plurality of nodes within said network, substantially without further user input.

6. The apparatus of claim 5, wherein:

the first user input is having a graphical representation of said network associated therewith; and

the method further comprises;

receiving second user input comprising one other single network description instruction; and

based at least in part on said second input, automatically updating, in accordance with said on other single instruction, said graphical representation of said network thereby effecting said RTE.

7. The apparatus of claim 5, wherein:

the first user input is having a graphical representation of said network associated therewith; and

the method further comprises;

receiving second user input comprising modifying said single network description instruction; and

based at least in part on said second input, automatically updating, in accordance with said modified single network description, said graphical representation of said network thereby effecting said RTE.

8. The apparatus of claim 5, wherein said plurality nodes is characterized by a probability distribution having a parameter associated therewith.

9. The apparatus of claim 8, wherein said parameter comprises spatial extent within said network, and generation of said plurality of nodes comprises distributing said plurality nodes within said network in accordance with said probability distribution with said spatial extent.

10. The apparatus of claim 5, wherein said single instruction comprises high-level network description (HLND) instruction comprising a keyword selected from the group consisting of CREATE, MAKE, and PUT.

11. A computer readable apparatus comprising a non-transitory storage medium adapted to store a computer program, said computer program which, when executed, causes creation of a plurality of probabilistic connections in a neural network comprising a plurality of nodes by:

based at least in part on receiving user input via a graphical user interface, automatically generating a single network description instruction, said single instructions configured to generate said plurality of probabilistic connections.

12. The apparatus of claim 11, wherein each of said plurality of probabilistic connections comprises one of (i) synapse;

and (ii) junction.

13. The apparatus of claim 11, wherein said plurality of probabilistic connections comprises connections between first portion of nodes and second portion of nodes of said plurality of nodes;

and

nodes of said first portion are being randomly selected from said plurality of nodes.

14. The apparatus of claim 13, wherein computer program is further configured to:

based at least in part on said user input, generate first representation of the network; and

based at least in part on said single network description instruction, generate second representation of the network;

wherein, said first representation matches said second representation thereby effecting round-trip engineering of said network.

15. The apparatus of claim 14, wherein said first representation comprises representation of said first portion and said second portion;

and

said second representation comprises, at least a part of, said single network description instruction.

16. The apparatus of claim 13, wherein said randomly selected nodes are characterized by a probability distribution having a parameter associated therewith.

17. The apparatus of claim 16, wherein parameter comprises an alphanumeric tag associated with nodes of said network.

18. The apparatus of claim 16, wherein parameter comprises a spatial coordinate associated with nodes of said network.

19. The apparatus of claim 11, wherein said plurality of probabilistic connections comprises connections selected randomly from all possible/attainable connections between first portion of nodes and second portion of nodes of said plurality of nodes.

20. Computer readable apparatus comprising a non-transitory storage medium, said storage medium comprising a plurality of instructions, said plurality of instructions comprising instructions configured to implement round-trip engineering of a neural network, comprising a plurality of nodes, by:

receive, via a graphical user interface, user input indicative of a portion of said plurality of nodes; and

based at least in part on said input, generate a single network description instruction configured to assign a tag to said portion.

21. The apparatus of claim 20, wherein assigning said tag to said portion comprises assigning said tag to each node of said portion.

22. The apparatus of claim 20, wherein:

said user input corresponds to first representation of the network;

said single network description instruction corresponds to second representation of the network; and

said first representation matches said second representation thereby effecting said round-trip engineering.

23. The apparatus of claim 20, wherein said tag comprises a sting identifier.

24. The apparatus of claim 20, wherein said tag comprises an alphanumeric identifier configured to identify said portion.

25. The apparatus of claim 24, wherein said alphanumeric identifier is adapted to indicate spatial coordinate of nodes within said portion.

26. The apparatus of claim 20, wherein said single instruction is configured consistent with English language structure and grammar.

27. The apparatus of claim 20, further comprising executing said single instruction by a computerized apparatus thereby assigning said tag to each node of said portion;

wherein said portion is characterized by a probability distribution, comprising a parameter indicative of a spatial coordinate; and

nodes of said portion are being distributed within said network in accordance with said probability distribution.

28. A system configured to implement round-trip engineering in a neural network, the system comprising:

one or more processors configured to execute one or more computer program modules, wherein execution of individual ones of the one or more computer program modules causes the one or more processors to;

receive a first user input indicative of generation of a plurality of nodes within the neural network; and

generate a single network description instruction based on the first user input, the single network description instruction being executable to cause generation of the plurality of nodes within the neural network substantially without further user input.

29. A system configured to implement round-trip engineering in a neural network, the system comprising:

one or more processors configured to execute one or more computer program modules, wherein execution of individual ones of the one or more computer program modules causes the one or more processors to;

receive, via a graphical user interface, user input indicative of a portion of the plurality of nodes; and

generate a single network description instruction based on the user input, the single network description instruction being configured to assign a tag to the portion.