Automated transfer of neural network definitions among federated areas

• US 10,002,029 B1
• Filed: 02/14/2018
• Issued: 06/19/2018
• Est. Priority Date: 02/05/2016
• Status: Active Grant

• Current Assignee: SAS Institute Incorporated
• Original Assignee: SAS Institute Incorporated
• Inventors: Bequet, Henry Gabriel Victor, Chen, Huina
• Primary Examiner(s): Patel, Hiren
• Application Number:
  US15/896,613
• Publication Number:
  US 20180173572A1
• Time in Patent Office: 125 Days
• Field of Search: None
• CPC Class Codes: G06F 16/9014 hash tables
• CPC Class Codes: G06F 16/90344 by using string matching techniques
• CPC Class Codes: G06F 9/5083 Techniques for rebalancing the load in a distributed system
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: G06N 5/01 Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound

Abstract

An apparatus includes a processor to: receive, from a first remote device, a request to perform iterations of a training job flow to generate a neural network at least partly within a first federated area, wherein access to the first federated area is granted to the first remote device and not a second remote device, access to a second federated area is granted to the second remote device and not the first remote device, and a transfer area is maintained to transfer a neural network data set between the first and second federated areas; perform the at least some iterations; and analyze an output object generated in each iteration to determine whether a condition has been met to transfer a copy of the neural network data set from the first federated area to the transfer area to enable its transfer to the second federated area to test the neural network.

Claims

1. An apparatus comprising a first processor and a first storage to store first instructions that, when executed by the first processor, cause the first processor to perform operations comprising:

receive, from a first remote device via a network, a first request to perform multiple iterations of a training job flow to generate a first neural network at least partly within a first federated area of a set of related federated areas, wherein;

the first federated area is maintained within one or more storage devices to store a training routine of the training job flow and a training data set;

the training data set comprises multiple sets of input values and corresponding output values representative of a function that the first neural network is to be trained to perform;

access via the network to the first federated area is granted to the first remote device and is denied to a second remote device;

a second federated area of the set of related federated areas is maintained within the one or more storage devices to store a testing routine of a testing job flow and a testing data set;

the testing data set comprises multiple sets of input values and corresponding output values to use in testing the performance of the function by the first neural network;

access via the network to the second federated area is granted to the second remote device and is denied to the first remote device;

a first transfer area is maintained within the one or more storage devices to support a transfer of a first neural network data set between the first federated area and the second federated area; and

the first neural network data set comprises a set of weight values and bias values that define the first neural network;

perform at least a subset of the multiple iterations of the training job flow at least partly within the first federated area, wherein each iteration of performance of the training job flow comprises an iteration of execution, by the first processor, of instructions of the training routine to generate at least a portion of the first neural network data set;

in response to each iteration of performance of the training job flow, analyze a first output object of the training routine to determine, by the first processor, whether a first predetermined condition has been met through the performance of at least the subset of the multiple iterations of the training job flow, wherein the first output object is selected from a group consisting of the first neural network data set and a first result report indicative of an extent of success of training of the first neural network;

in response to the first predetermined condition having been met, transfer a copy of the first neural network data set from the first federated area to the first transfer area to enable the first neural network data set to be transferred to the second federated area to enable multiple iterations of a testing job flow to be performed with the first neural network data set to test the first neural network at least partly within the second federated area, wherein, in response to each iteration of performance of the testing job flow, an iteration of a second output object that is output during the iteration of performance of the testing job flow is analyzed to determine whether the first neural network does successfully perform the function with at least the testing data set; and

in response to at least the determination that the first neural network does successfully perform the function with the testing data set, provide access, to another device via the network, to a representation of the first neural network, to enable use of the first neural network by the other device.

2. The apparatus of claim 1, comprising a second processor and a second storage to store second instructions that, when executed by the second processor, cause the second processor to perform operations comprising:

monitor the first transfer area for the transfer of the first neural network data set into the first transfer area;

in response to the transfer of the first neural network data set into the first transfer area, transfer the first neural network data set into the second federated area; and

in response to the transfer of the first neural network data set into the second federated area, perform at least a subset of the multiple iterations of the testing job flow with the first neural network data set at least partly within the second federated area to test an extent of success of performance of the function by the first neural network.

3. The apparatus of claim 2, wherein:

each iteration of performance of the testing job flow comprises an iteration of execution, by the second processor, of instructions of a testing routine;

each iteration of execution of the instructions of the testing routine generates an iteration of the second output object indicative of the extent of success of performance of the function by the first neural network; and

the second processor is caused to perform operations comprising;

in response to each iteration of performance of the testing job flow, analyze the second output object to determine, by the second processor, whether a second predetermined condition has been met to determine that the first neural network does not successfully perform the function; and

in response to the second predetermined condition having been met, transfer a copy of the second output object from the second federated area to the first transfer area to enable the second output object to be transferred to the first federated area.

4. The apparatus of claim 2, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of the first neural network data set between the second federated area and the third federated area;

each iteration of performance of the testing job flow comprises an iteration of execution, by the second processor, of instructions of a testing routine;

each iteration of execution of the instructions of the testing routine generates an iteration of the second output object indicative of the extent of success of performance of the function by the first neural network; and

the second processor is caused to perform operations comprising;

in response to each iteration of performance of the testing job flow, analyze the second output object to determine, by the second processor, whether a third predetermined condition has been met to determine that the first neural network does successfully perform the function; and

in response to the third predetermined condition having been met, transfer a copy of the first neural network data set from the second federated area to the second transfer area to enable the first neural network data set to be transferred to the third federated area.

5. The apparatus of claim 2, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of a second neural network data set between the second federated area and the third federated area;

the second neural network data set comprises a set of weight values and bias values that define a second neural network comprising a refined version of the first neural network;

each iteration of performance of the testing job flow comprises an iteration of execution, by the second processor, of instructions of a testing routine;

each iteration of execution of the instructions of the testing routine generates at least a portion of the second neural network data set based on the first neural network data set and the extent of success of performance of the function by the first neural network during the iteration; and

the second processor is caused to perform operations comprising;

in response to each iteration of performance of the testing job flow, analyze the second output object of the testing routine to determine, by the second processor, whether a fourth predetermined condition has been met through the performance of at least the subset of the multiple iterations of the testing job flow, wherein the second output object is selected from a group consisting of the second neural network data set and a second result report indicative of an extent of success of performance of the function by the neural network; and

in response to the fourth predetermined condition having been met, transfer a copy of the second neural network data set from the second federated area to the second transfer area to enable the second neural network data set to be transferred to the third federated area.

6. The apparatus of claim 1, wherein:

the set of federated areas comprises a hierarchical tree of federated areas;

the tree comprises a base federated area at a root position of the tree;

the first federated area is positioned within a first branch of the tree;

the second federated area is positioned within a second branch of the tree separate from the first branch; and

the first transfer area is instantiated within the base federated area, or within a storage area within the one or more storage devices at which the first federated area and the second federated area overlap.

7. The apparatus of claim 6, wherein:

the first processor is caused to perform operations comprising;

maintain a first inheritance relationship between the first federated area and the base federated area in which a base object stored within the base federated area is made as accessible to the first remote device as a first object stored within the first federated area; and

maintain a first priority relationship between the first federated area and the base federated area as an exception to the first inheritance relationship in which the first processor is caused to;

search within the first federated area for a first required object that is required to enable the performance of the training job flow;

in response to locating the first required object within the first federated area, refrain from searching the base federated area for the first required object; and

in response to failing to locate the first required object within the first federated area, search within the base federated area for the first required object; and

the apparatus comprises a second processor and a second storage to store second instructions that, when executed by the second processor, cause the second processor to perform operations comprising;

maintain a second inheritance relationship between the second federated area and the base federated area in which the base object stored within the base federated area is made as accessible to the second remote device as a second object stored within the second federated area; and

maintain a second priority relationship between the second federated area and the base federated area as an exception to the second inheritance relationship in which the second processor is caused to;

search within the second federated area for a second required object that is required to enable the performance of the testing job flow;

in response to locating the second required object within the second federated area, refrain from searching the base federated area for the second required object; and

in response to failing to locate the second required object within the second federated area, search within the base federated area for the second required object.

8. The apparatus of claim 6, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of either the first neural network data set or a second neural network data set derived from the first neural network data set between the second federated area and the third federated area;

the third federated area is positioned within a third branch of the tree separate from the first branch and the second branch; and

the second transfer area is instantiated within the base federated area, or within a storage area within the one or more storage devices at which the second federated area and the third federated area overlap.

9. The apparatus of claim 8, wherein aspects of the transfer of the first neural network data set from the first federated area to the second federated area through the first transfer area, and aspects of the transfer of either the first neural network data set or the second neural network data set from the second federated area to the third federated area, are defined as parts of a transfer flow by a transfer flow definition stored in the base federated area.

10. The apparatus of claim 9, wherein the first processor is caused to perform operations comprising:

receive, via the network, from one of the first remote device, the second remote device and the third remote device, a second request to instantiate the transfer flow among the first federated area, the second federated area and the third federated area;

examine a first security credential associated with the first remote device to determine whether the second request is authorized for the first federated area;

examine a second security credential associated with the second remote device to determine whether the second request is authorized for the second federated area;

examine a third security credential associated with the second remote device to determine whether the second request is authorized for the third federated area;

in response to a determination that the second request is authorized by the first remote device, the second remote device and the third remote device, the first processor is caused to perform operations comprising;

instantiate the first transfer area;

instantiate the second transfer area;

store, within the first federated area or within the base federated area, a first transfer routine comprising instructions to cause the analysis of the first output object and the transfer of the first neural network data set from the first federated area to at least the first transfer area; and

store, within the second federated area or within the base federated area, a second transfer routine comprising instructions to cause an analysis of the second output object indicative of an extent of success of performance of the function by the first neural network, and the transfer of either the first neural network data set or the second neural network data set from the second federated area to at least the second transfer area based on the analysis of the second output object.

11. A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, the computer-program product including first instructions operable to cause a first processor to perform operations comprising:

receive, from a first remote device via a network, a first request to perform multiple iterations of a training job flow to generate a first neural network at least partly within a first federated area of a set of related federated areas, wherein;

the first federated area is maintained within one or more storage devices to store a training routine of the training job flow and a training data set;

the training data set comprises multiple sets of input values and corresponding output values representative of a function that the first neural network is to be trained to perform;

access via the network to the first federated area is granted to the first remote device and is denied to a second remote device;

a second federated area of the set of related federated areas is maintained within the one or more storage devices to store a testing routine of a testing job flow and a testing data set;

the testing data set comprises multiple sets of input values and corresponding output values to use in testing the performance of the function by the first neural network;

access via the network to the second federated area is granted to the second remote device and is denied to the first remote device;

a first transfer area is maintained within the one or more storage devices to support a transfer of a first neural network data set between the first federated area and the second federated area; and

the first neural network data set comprises a set of weight values and bias values that define the first neural network;

perform at least a subset of the multiple iterations of the training job flow at least partly within the first federated area, wherein each iteration of performance of the training job flow comprises an iteration of execution, by the first processor, of instructions of the training routine to generate at least a portion of the first neural network data set;

in response to each iteration of performance of the training job flow, analyze a first output object of the training routine to determine, by the first processor, whether a first predetermined condition has been met through the performance of at least the subset of the multiple iterations of the training job flow, wherein the first output object is selected from a group consisting of the first neural network data set and a first result report indicative of an extent of success of training of the first neural network;

in response to the first predetermined condition having been met, transfer a copy of the first neural network data set from the first federated area to the first transfer area to enable the first neural network data set to be transferred to the second federated area to enable multiple iterations of a testing job flow to be performed with the first neural network data set to test the first neural network at least partly within the second federated area, wherein, in response to each iteration of performance of the testing job flow, an iteration of a second output object that is output during the iteration of performance of the testing job flow is analyzed to determine whether the first neural network does successfully perform the function with at least the testing data set; and

in response to at least the determination that the first neural network does successfully perform the function with the testing data set, provide access, to another device via the network, to a representation of the first neural network, to enable use of the first neural network by the other device.

12. The computer-program product of claim 11, the computer-program product including second instructions operable to cause a second processor to perform operations comprising:

monitor the first transfer area for the transfer of the first neural network data set into the first transfer area;

in response to the transfer of the first neural network data set into the first transfer area, transfer the first neural network data set into the second federated area; and

in response to the transfer of the first neural network data set into the second federated area, perform at least a subset of the multiple iterations of the testing job flow with the first neural network data set at least partly within the second federated area to test an extent of success of performance of the function by the first neural network.

13. The computer-program product of claim 12, wherein:

each iteration of performance of the testing job flow comprises an iteration of execution, by the second processor, of instructions of a testing routine;

each iteration of execution of the instructions of the testing routine generates an iteration of the second output object indicative of the extent of success of performance of the function by the first neural network; and

the second processor is caused to perform operations comprising;

in response to each iteration of performance of the testing job flow, analyze the second output object to determine, by the second processor, whether a second predetermined condition has been met to determine that the first neural network does not successfully perform the function; and

in response to the second predetermined condition having been met, transfer a copy of the second output object from the second federated area to the first transfer area to enable the second output object to be transferred to the first federated area.

14. The computer-program product of claim 12, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of the first neural network data set between the second federated area and the third federated area;

each iteration of performance of the testing job flow comprises an iteration of execution, by the second processor, of instructions of a testing routine;

each iteration of execution of the instructions of the testing routine generates an iteration of the second output object indicative of the extent of success of performance of the function by the first neural network; and

the second processor is caused to perform operations comprising;

in response to each iteration of performance of the testing job flow, analyze the second output object to determine, by the second processor, whether a third predetermined condition has been met to determine that the first neural network does successfully perform the function; and

in response to the third predetermined condition having been met, transfer a copy of the first neural network data set from the second federated area to the second transfer area to enable the first neural network data set to be transferred to the third federated area.

15. The computer-program product of claim 12, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of a second neural network data set between the second federated area and the third federated area;

the second neural network data set comprises a set of weight values and bias values that define a second neural network comprising a refined version of the first neural network;

each iteration of performance of the testing job flow comprises an iteration of execution, by the second processor, of instructions of a testing routine;

each iteration of execution of the instructions of the testing routine generates at least a portion of the second neural network data set based on the first neural network data set and the extent of success of performance of the function by the first neural network during the iteration; and

the second processor is caused to perform operations comprising;

in response to each iteration of performance of the testing job flow, analyze the second output object of the testing routine to determine, by the second processor, whether a fourth predetermined condition has been met through the performance of at least the subset of the multiple iterations of the testing job flow, wherein the second output object is selected from a group consisting of the second neural network data set and a second result report indicative of an extent of success of performance of the function by the neural network; and

in response to the fourth predetermined condition having been met, transfer a copy of the second neural network data set from the second federated area to the second transfer area to enable the second neural network data set to be transferred to the third federated area.

16. The computer-program product of claim 11, wherein:

the set of federated areas comprises a hierarchical tree of federated areas;

the tree comprises a base federated area at a root position of the tree;

the first federated area is positioned within a first branch of the tree;

the second federated area is positioned within a second branch of the tree separate from the first branch; and

the first transfer area is instantiated within the base federated area, or within a storage area within the one or more storage devices at which the first federated area and the second federated area overlap.

17. The computer-program product of claim 16, wherein:

the first processor is caused to perform operations comprising;

maintain a first inheritance relationship between the first federated area and the base federated area in which a base object stored within the base federated area is made as accessible to the first remote device as a first object stored within the first federated area; and

maintain a first priority relationship between the first federated area and the base federated area as an exception to the first inheritance relationship in which the first processor is caused to;

search within the first federated area for a first required object that is required to enable the performance of the training job flow;

in response to locating the first required object within the first federated area, refrain from searching the base federated area for the first required object; and

in response to failing to locate the first required object within the first federated area, search within the base federated area for the first required object; and

the computer-program product includes second instructions operable to cause a second processor to perform operations comprising;

maintain a second inheritance relationship between the second federated area and the base federated area in which the base object stored within the base federated area is made as accessible to the second remote device as a second object stored within the second federated area; and

maintain a second priority relationship between the second federated area and the base federated area as an exception to the second inheritance relationship in which the second processor is caused to;

search within the second federated area for a second required object that is required to enable the performance of the testing job flow;

in response to locating the second required object within the second federated area, refrain from searching the base federated area for the second required object; and

in response to failing to locate the second required object within the second federated area, search within the base federated area for the second required object.

18. The computer-program product of claim 16, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of either the first neural network data set or a second neural network data set derived from the first neural network data set between the second federated area and the third federated area;

the third federated area is positioned within a third branch of the tree separate from the first branch and the second branch; and

the second transfer area is instantiated within the base federated area, or within a storage area within the one or more storage devices at which the second federated area and the third federated area overlap.

19. The computer-program product of claim 18, wherein aspects of the transfer of the first neural network data set from the first federated area to the second federated area through the first transfer area, and aspects of the transfer of either the first neural network data set or the second neural network data set from the second federated area to the third federated area, are defined as parts of a transfer flow by a transfer flow definition stored in the base federated area.

20. The computer-program product of claim 19, wherein the first processor is caused to perform operations comprising:

receive, via the network, from one of the first remote device, the second remote device and the third remote device, a second request to instantiate the transfer flow among the first federated area, the second federated area and the third federated area;

examine a first security credential associated with the first remote device to determine whether the second request is authorized for the first federated area;

examine a second security credential associated with the second remote device to determine whether the second request is authorized for the second federated area;

examine a third security credential associated with the second remote device to determine whether the second request is authorized for the third federated area;

in response to a determination that the second request is authorized by the first remote device, the second remote device and the third remote device, the first processor is caused to perform operations comprising;

instantiate the first transfer area;

instantiate the second transfer area;

store, within the first federated area or within the base federated area, a first transfer routine comprising instructions to cause the analysis of the first output object and the transfer of the first neural network data set from the first federated area to at least the first transfer area; and

store, within the second federated area or within the base federated area, a second transfer routine comprising instructions to cause an analysis of the second output object indicative of an extent of success of performance of the function by the first neural network, and the transfer of either the first neural network data set or the second neural network data set from the second federated area to at least the second transfer area based on the analysis of the second output object.

21. A computer-implemented method comprising:

receiving, from a first remote device via a network, a first request to perform multiple iterations of a training job flow to generate a first neural network at least partly within a first federated area of a set of related federated areas, wherein;

the first federated area is maintained within one or more storage devices to store a training routine of the training job flow and a training data set;

the training data set comprises multiple sets of input values and corresponding output values representative of a function that the first neural network is to be trained to perform;

access via the network to the first federated area is granted to the first remote device and is denied to a second remote device;

a second federated area of the set of related federated areas is maintained within the one or more storage devices to store a testing routine of a testing job flow and a testing data set;

the testing data set comprises multiple sets of input values and corresponding output values to use in testing the performance of the function by the first neural network;

access via the network to the second federated area is granted to the second remote device and is denied to the first remote device;

a first transfer area is maintained within the one or more storage devices to support a transfer of a first neural network data set between the first federated area and the second federated area; and

the first neural network data set comprises a set of weight values and bias values that define the first neural network;

performing at least a subset of the multiple iterations of the training job flow at least partly within the first federated area, wherein each iteration of performance of the training job flow comprises an iteration of executing instructions of the training routine, by a first processor, to generate at least a portion of the first neural network data set;

in response to each iteration performing the training job flow, analyzing a first output object of the training routine to determine, by the first processor, whether a first predetermined condition has been met through the performance of at least the subset of the multiple iterations of the training job flow, wherein the first output object is selected from a group consisting of the first neural network data set and a first result report indicative of an extent of success of training of the first neural network;

in response to the first predetermined condition having been met, transferring a copy of the first neural network data set from the first federated area to the first transfer area to enable the first neural network data set to be transferred to the second federated area to enable multiple iterations of a testing job flow to be performed with the first neural network data set to test the first neural network at least partly within the second federated area, wherein, in response to each iteration of performance of the testing job flow, an iteration of a second output object that is output during the iteration of performance of the testing job flow is analyzed to determine whether the first neural network does successfully perform the function with at least the testing data set; and

in response to at least the determination that the first neural network does successfully perform the function with the testing data set, providing access, to another device via the network, to a representation of the first neural network, to enable use of the first neural network by the other device.

22. The computer-implemented method of claim 21, comprising:

monitoring the first transfer area for the transfer of the first neural network data set into the first transfer area;

in response to the transfer of the first neural network data set into the first transfer area, transferring the first neural network data set into the second federated area; and

in response to the transfer of the first neural network data set into the second federated area, performing at least a subset of the multiple iterations of the testing job flow with the first neural network data set at least partly within the second federated area to test an extent of success of performance of the function by the first neural network.

23. The computer-implemented method of claim 22, wherein:

each iteration of performance of the testing job flow comprises an iteration of executing instructions of a testing routine, by a second processor;

each iteration of executing the instructions of the testing routine comprises generating an iteration of the second output object indicative of the extent of success of performance of the function by the first neural network; and

the method comprises;

in response to each iteration of performance of the testing job flow, analyzing the second output object to determine, by the second processor, whether a second predetermined condition has been met to determine that the first neural network does not successfully perform the function; and

in response to the second predetermined condition having been met, transferring a copy of the second output object from the second federated area to the first transfer area to enable the second output object to be transferred to the first federated area.

24. The computer-implemented method of claim 22, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of the first neural network data set between the second federated area and the third federated area;

each iteration of performance of the testing job flow comprises an iteration of executing instructions of a testing routine by a second processor;

each iteration of executing the instructions of the testing routine comprises generating an iteration of the second output object indicative of the extent of success of performance of the function by the first neural network; and

the method comprises;

in response to each iteration of performance of the testing job flow, analyzing the second output object to determine, by the second processor, whether a third predetermined condition has been met to determine that the first neural network does successfully perform the function; and

in response to the third predetermined condition having been met, transferring a copy of the first neural network data set from the second federated area to the second transfer area to enable the first neural network data set to be transferred to the third federated area.

25. The computer-implemented method of claim 22, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of a second neural network data set between the second federated area and the third federated area;

the second neural network data set comprises a set of weight values and bias values that define a second neural network comprising a refined version of the first neural network;

each iteration of performance of the testing job flow comprises an iteration of executing instructions of a testing routine by a second processor;

each iteration of executing the instructions of the testing routine comprises generating at least a portion of the second neural network data set based on the first neural network data set and the extent of success of performance of the function by the first neural network during the iteration; and

the method comprises;

in response to each iteration of performance of the testing job flow, analyzing the second output object of the testing routine to determine, by the second processor, whether a fourth predetermined condition has been met through the performance of at least the subset of the multiple iterations of the testing job flow, wherein the second output object is selected from a group consisting of the second neural network data set and a second result report indicative of an extent of success of performance of the function by the neural network; and

in response to the fourth predetermined condition having been met, transferring a copy of the second neural network data set from the second federated area to the second transfer area to enable the second neural network data set to be transferred to the third federated area.

26. The computer-implemented method of claim 21, wherein:

the set of federated areas comprises a hierarchical tree of federated areas;

the tree comprises a base federated area at a root position of the tree;

the first federated area is positioned within a first branch of the tree;

the second federated area is positioned within a second branch of the tree separate from the first branch; and

the first transfer area is instantiated within the base federated area, or within a storage area within the one or more storage devices at which the first federated area and the second federated area overlap.

27. The computer-implemented method of claim 26, comprising:

maintaining a first inheritance relationship between the first federated area and the base federated area in which a base object stored within the base federated area is made as accessible to the first remote device as a first object stored within the first federated area;

maintaining a first priority relationship between the first federated area and the base federated area as an exception to the first inheritance relationship by performing operations comprising;

searching within the first federated area for a first required object that is required to enable the performance of the training job flow;

in response to locating the first required object within the first federated area, refraining from searching the base federated area for the first required object; and

in response to failing to locate the first required object within the first federated area, searching within the base federated area for the first required object;

maintaining a second inheritance relationship between the second federated area and the base federated area in which the base object stored within the base federated area is made as accessible to the second remote device as a second object stored within the second federated area; and

maintaining a second priority relationship between the second federated area and the base federated area as an exception to the second inheritance relationship by performing operations comprising;

searching within the second federated area for a second required object that is required to enable the performance of the testing job flow;

in response to locating the second required object within the second federated area, refraining from searching the base federated area for the second required object; and

in response to failing to locate the second required object within the second federated area, searching within the base federated area for the second required object.

28. The computer-implemented method of claim 26, wherein:

a third federated area of the set of related federated areas is maintained within the one or more storage devices to store an experimental use routine of an experimental use job flow and an experimental use data set;

the experimental use data set comprises multiple sets of input values and corresponding output values to use in further testing the performance of the function by the first neural network in multiple experimental use scenarios;

access via the network to the third federated area is granted to a third remote device, and is denied to the first remote device and the second remote device;

access via the network to the first federated area and the second federated area is denied to the third remote device;

a second transfer area is maintained within the one or more storage devices to support a transfer of either the first neural network data set or a second neural network data set derived from the first neural network data set between the second federated area and the third federated area;

the third federated area is positioned within a third branch of the tree separate from the first branch and the second branch; and

the second transfer area is instantiated within the base federated area, or within a storage area within the one or more storage devices at which the second federated area and the third federated area overlap.

29. The computer-implemented method of claim 28, wherein aspects of the transfer of the first neural network data set from the first federated area to the second federated area through the first transfer area, and aspects of the transfer of either the first neural network data set or the second neural network data set from the second federated area to the third federated area, are defined as parts of a transfer flow by a transfer flow definition stored in the base federated area.

30. The computer-implemented method of claim 29, comprising:

receiving, via the network, from one of the first remote device, the second remote device and the third remote device, a second request to instantiate the transfer flow among the first federated area, the second federated area and the third federated area;

examining a first security credential associated with the first remote device to determine whether the second request is authorized for the first federated area;

examining a second security credential associated with the second remote device to determine whether the second request is authorized for the second federated area;

examining a third security credential associated with the second remote device to determine whether the second request is authorized for the third federated area;

in response to a determination that the second request is authorized by the first remote device, the second remote device and the third remote device, performing operations comprising;

instantiating the first transfer area;

instantiating the second transfer area;

storing, within the first federated area or within the base federated area, a first transfer routine comprising instructions to cause the analysis of the first output object and the transfer of the first neural network data set from the first federated area to at least the first transfer area; and

storing, within the second federated area or within the base federated area, a second transfer routine comprising instructions to cause an analysis of the second output object indicative of an extent of success of performance of the function by the first neural network, and the transfer of either the first neural network data set or the second neural network data set from the second federated area to at least the second transfer area based on the analysis of the second output object.