[미국특허]
High level neuromorphic network description apparatus and methods
원문보기
IPC분류정보
국가/구분
United States(US) Patent
공개
국제특허분류(IPC7판)
G06N-003/02
G06N-003/04
출원번호
US-0385933
(2012-03-15)
공개번호
US-0073500
(2013-03-21)
발명자
/ 주소
Szatmary, Botond
Izhikevich, Eugene
출원인 / 주소
Szatmary, Botond
인용정보
피인용 횟수 :
0인용 특허 :
0
초록▼
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 t
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.
대표청구항▼
1. A computer realized method of implementing a neural network using an instruction set, the method comprising: providing a representation of the neural network, the representation comprising a plurality of instructions of said instruction set; andcompiling said representation into to machine execut
1. A computer realized method of implementing a neural network using an instruction set, the method comprising: providing a representation of the neural network, the representation comprising a plurality of instructions of said instruction set; andcompiling said representation into to machine executable format;wherein the instruction set comprises a structured language configured consistent with English language structure and grammar. 2. The method of claim 1, wherein said instruction set comprises a first instruction adapted to cause generation of at least one node within said network. 3. The method of claim 2, wherein said first instruction comprises a keyword selected from the group consisting of CREATE, MAKE, and PUT. 4. The method of claim 3, wherein said keyword comprises only one keyword. 5. The method of claim 2, wherein: said at least one node is characterized by a spatial coordinate; andsaid first instruction comprises a structured English language statement configured to assign a desired location within said network to said spatial coordinate, said assignment being performed as a part of said generation of said at least one node. 6. The method of claim 5, wherein said first instruction comprises a keyword selected from the group consisting of ON, IN, and AT. 7. The method of claim 6, wherein said keyword comprises only one keyword. 8. The method of claim 2, wherein: said generation of at least one node comprises generation of a first node and a second node; andthe instruction set comprises a second instruction configured to cause a connection between said first node and said second node. 9. The method of claim 8, wherein said connection comprises a synapse. 10. The method of claim 8, wherein said connection comprises a junction. 11. The method of claim 8, wherein: said generation of at least one node further comprises generation of plurality of nodes, the plurality comprising the first node and the second node; andsaid second instruction comprises a Boolean expression configured to select two subsets within said plurality of nodes and to cause a plurality of connection between nodes of one of said two subsets and nodes of another of said two subsets. 12. The method of claim 11, wherein said second instruction comprises a keyword selected from the group consisting of CONNECT, LINK, and PROJECT. 13. The method of claim 11, wherein said Boolean expression comprises a keyword selected from the group consisting of AND, NOT, and OR. 14. The method of claim 8, wherein the second instruction comprises a keyword selected from the group consisting of WITH, BY, and USING, said keyword adapted to specify a type of said connection. 15. A method of programming a computerized apparatus, comprising: representing a neural network using a plurality of instructions of an instruction set; andcompiling said representation into to machine representation for execution by said computerized apparatus;wherein the instruction set comprises a structured language configured consistent with English language structure and grammar. 16. The method of claim 15, wherein: said network comprises a plurality of elements, each element of said plurality of elements having a tag associated therewith; andsaid instruction set comprises a first instruction configured to identify a subset of said plurality of elements, based at least in part on said tag. 17. The method of claim 16, wherein said compiling is effected by a database configured to store a plurality of tags, said plurality of tags comprising the tag. 18. The method of claim 16, wherein said instruction set comprises a second instruction configured to effect assignment of a new tag to said subset. 19. The method of claim 18, wherein said second instruction comprises a keyword selected from the list consisting of TAG, ASSIGN, and MARK. 20. The method of claim 15, wherein said instruction set further comprises a Boolean expression configured consistent with a structured English representation; and said structured English representation enables machine execution of said Boolean expression by at least in part an implicit assignment of logical AND operation between any two Boolean variables of said Boolean expression, where said Boolean variables are separated by a separator keyword thereby. 21. The method of claim 20, wherein said separator keyword comprises a whitespace. 22. A method of operating a neuromorphic computerized apparatus comprising a nonvolatile storage medium storing an instruction set, compiling kernel, and processing module, the method comprising: providing a representation of a neural network to the processing module, the representation comprising a plurality of instructions of said instruction set; andencoding said representation into hardware independent format by the processing module using said kernel;wherein the instruction set comprises a structured language configured consistent with English language structure and grammar. 23. The method of claim 22, wherein said hardware independent format is configured to enable conversion of said representation into a plurality of machine executable instructions; and said plurality of machine executable instructions effecting operation of said neural network. 24. The method of claim 23, wherein said hardware independent format comprises Elementary Network Description (END) format. 25. The method of claim 24, wherein said machine executable instructions are selected from the group consisting of central processing unit (CPU) instructions, graphics processing unit (GPU) instructions, and field programmable gate array (FPGA) instructions. 26. The method of claim 23, wherein said machine format comprises a plurality of functions specifically configured for execution by said processing module. 27. The method of claim 23, wherein: said computerized apparatus comprises a speech input apparatus; andproviding said representation is effected by a user of said computerized apparatus, said user speaking said plurality of instructions. 28. The method of claim 27, wherein speaking said plurality of instructions is effected one instruction at a time. 29. The method of claim 27, wherein: said speech input apparatus is operably coupled to said kernel; andsaid kernel comprises a speech-recognition module configured to translate digitized user speech into one or more instructions of said instruction set. 30. The method of claim 23, wherein said instruction set comprises a first instruction configured to effect generation of at least one node within said network. 31. The method of claim 30, wherein: said at least one node is characterized by a spatial coordinate; andsaid first instruction comprises a structured English language statement configured to assign to a desired location within said network to said spatial coordinate, said assignment being effected substantially during creation of said at least one node. 32. The method of claim 30, wherein said generation of said at least one node comprises generation of a first node and a second node; and the instruction set comprises a second instruction configured to effect a connection between said first node and said second node. 33. A system configured to implement a neural network using an instruction set, 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: provide a representation of the neural network, the representation comprising a plurality of instructions of the instruction set, the instruction set comprising a structured language configured consistent with English language structure and grammar; andcompile the representation into to machine executable format.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.