Compositional prototypes for scalable neurosynaptic networks

Patent metadata

Field	Value
Publication number	US-2016283840-A1
Application number	US-201514669575-A
Country	US
Kind code	A1
Filing date	Mar 26, 2015
Priority date	Mar 26, 2015
Publication date	Sep 29, 2016
Grant date	—

Abstract

Embodiments of the invention provide a method comprising maintaining a library of one or more compositional prototypes. Each compositional prototype is associated with a neurosynaptic program. The method further comprises searching the library based on one or more search parameters. At least one compositional prototype satisfying the search parameters is selected. A neurosynaptic network is generated or extended by applying one or more rules associated with the selected compositional prototypes.

First claim

What is claimed is:

1 . A method, comprising:
maintaining a library of one or more compositional prototypes, wherein each compositional prototype is associated with one or more neurosynaptic programs; searching the library based on one or more search parameters; selecting at least one compositional prototype satisfying the search parameters; and generating or extending a neurosynaptic network by applying one or more rules associated with the selected compositional prototypes.

2 . The method of claim 1 , wherein the rules control one or more of the following: adding a neurosynaptic program to the network, adding a sub-network to the network, adding a connection in the network, adding a neuron in the network, changing one or more parameters of the network, applying a core prototype, and enhancing functionality of the network.

3 . The method of claim 1 , wherein the compositional prototypes maintained include at least one of the following: a compositional corelet prototype, a core prototype, a synaptic crossbar prototype, a network connectivity permutation, and spike coding transcoding.

4 . The method of claim 1 , wherein the search parameters include at least one of the following: a name of a neurosynaptic program, a core circuit functionality, a network topology, a spike coding scheme, and a network size.

5 . The method of claim 1 , where the search parameters are selected or automatically computed from a given network, and the rules are applied to generate, extend, enhance or repair the same network.

6 . The method of claim 1 , wherein:
at least one compositional prototype satisfying the search parameters is selected based on user selection; and the rules are applied to instantiate one or more neurosynaptic programs for generating or extending the network.

7 . The method of claim 1 , wherein the neurosynaptic programs are instantiated in a distributed environment.

8 . The method of claim 1 , wherein the rules include instantiating a transducer corelet to interconnect two neurosynaptic programs operating on different spike coding schemes.

9 . The method of claim 1 , further comprising:
populating the library by automatically identifying one or more compositional prototypes based on one or more model files.

10 . A method, comprising:
analyzing one or more model files, wherein each model file includes information relating to a neurosynaptic core; identifying one or more unique core patterns in said model files; assigning each unique core pattern identified with a corresponding unique identifier; and for each unique core pattern identified:
locating each instance of said unique core pattern in said model files based on repetitions of a corresponding unique identifier for said unique core pattern; and
replacing each instance located with a compositional prototype corresponding to said unique core pattern.

11 . The method of claim 10 , wherein each unique identifier comprises a hash key.

12 . The method of claim 10 , wherein each unique core pattern identified relates to one of the following: a neurosynaptic core, a portion of a neurosynaptic core.

13 . The method of claim 10 , wherein each compositional prototype comprises one of the following: a core prototype, a crossbar prototype.

14 . The method of claim 10 , further comprising:
associating each unique core pattern identified with a corresponding parametric call to one of a function or a corelet for generating said unique core pattern.

15 . The method of claim 10 , wherein:
each unique core pattern identified relates to a connectivity permutation between two or more corelets.

16 . The method of claim 15 , further comprising:
analyzing said model files based on one or more algorithms; extracting metadata from each model file, wherein metadata extracted from each model file identifies a corelet that generated said model file; and based on metadata extracted from each model file, determining at least one connectivity permutation.

17 . A method, comprising:
providing a programming environment for corelet composition; recording, utilizing said programming environment, one or more user actions associated with corelet composition; for each user action recorded, maintaining a corresponding database record including metadata related to said user action; and clustering database records with similar metadata to identify one or more patterns related to corelet composition.

18 . The method of claim 17 , wherein:
each pattern identified relates to a connectivity permutation between two or more corelets; for each user action recorded, a corresponding database record for said user action includes metadata identifying each of the following: two or more corelets, and a connectivity permutation for connecting said two or more corelets; and database records with metadata identifying the same corelets are clustered together.

19 . The method of claim 17 , wherein:
each pattern identified relates to a compositional prototype.

Assignees

• IBM

Inventors

• Amir Arnon
• Datta Pallab
• Modha Dharmendra S
• Shaw Benjamin G

Classifications

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06F17/30507

  Physics · mapped topic

• G06N3/063 Primary

  using electronic means · CPC title

• G06N3/105

  Shells for specifying net layout · CPC title

• G06N3/082

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