Anti-fragile network

1. A computer-implemented method comprising:

receiving, at one or more computing devices, parameter data from a network of nodes, the parameter data comprising attributes, policies, and action spaces for each node included in the network of nodes;

configuring, by the one or more computing devices, one or more interruptive events on one or more nodes included in the network of nodes;

determining, by the one or more computing devices, a first action of each node included in the network of nodes in response to the one or more interruptive events;

determining, by the one or more computing devices, a first performance metric, for each node, that corresponds to the first action, wherein the first performance metric is determined based on at least a first reward value associated with the first action;

performing a reinforcement learning process to train a machine learning model to generate actions to be performed by nodes that improve anti-fragility of the network, including continuously updating, by the one or more computing devices, the first action for each node in the reinforcement learning process until an anti-fragility metric of a final action satisfies a performance threshold representing the anti-fragility of the network, wherein updating the first action for a node in the reinforcement learning process comprises:

determining an updated action for the node based on the first performance metric of the first action, and

determining an updated performance metric, for the node, that corresponds to the updated action, wherein the updated performance metric is determined based on an updated reward value associated with the updated action, and wherein the updated reward value is larger than the first reward value; and

transmitting, by the one or more computing devices, the final action for each node to the network of nodes, whereby each node performs the final action in response to the one or more interruptive events.

2. The computer-implemented method of claim 1 , wherein setting the one or more interruptive events comprises:

determining a probability distribution of the one or more interruptive events.

3. The computer-implemented method of claim 1 , wherein determining an action for each node in response to the one or more interruptive events comprises:

determining the action that are included in the action spaces of the node and in accordance with the attributes and the policies of the node.

4. The computer-implemented method of claim 1 , wherein determining an action for each node comprises:

determining the action using a simulation process.

5. The computer-implemented method of claim 1 , wherein a performance metric of each node comprises whether the node resolves the one or more interruptive events.

6. The computer-implemented method of claim 5 , wherein whether the node resolves the one or more interruptive events are determined based on success criteria that include whether a performance threshold is satisfied under a set of predetermined conditions.

7. The computer-implemented method of claim 6 , the set of predetermined conditions comprises a set of threshold values for various costs defined in the policies of the node.

8. The computer-implemented method of claim 5 , wherein the performance metric for each node further comprises a difference between i) a reward value of the node corresponding to an action in response to the one or more interruptive events and ii) a threshold reward value that indicates an average reward of the node without the one or more interruptive events.

9. The computer-implemented method of claim 5 , wherein the performance metric for each node further comprises a response time to the one or more interruptive events.

10. The computer-implemented method of claim 1 , comprising:

determining a performance metric, for the network of nodes, that includes a number of nodes that resolve the one or more interruptive events; or

determining the performance metric, for the network of nodes, that includes a percentage of nodes that resolve the one or more interruptive events.

11. The computer-implemented method of claim 1 , comprising:

configuring one or more second interruptive events in a curriculum learning process.

12. A non-transitory computer-readable medium encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform operations comprising:

receiving parameter data from a network of nodes, the parameter data comprising attributes, policies, and action spaces for each node included in the network of nodes;

configuring one or more interruptive events on one or more nodes included in the network of nodes;

determining a first action of each node included in the network of nodes in response to the one or more interruptive events;

determining a first performance metric, for each node, that corresponds to the first action, wherein the first performance metric is determined based on at least a first reward value associated with the first action;

performing a reinforcement learning process to train a machine learning model to generate actions to be performed by nodes that improve anti-fragility of the network, including continuously updating the first action for each node in the reinforcement learning process until an anti-fragility metric of a final action satisfies a performance threshold representing the anti-fragility of the network, wherein updating the first action for a node in the reinforcement learning process comprises:

determining an updated action for the node based on the first performance metric of the first action, and

determining an updated performance metric, for the node, that corresponds to the updated action, wherein the updated performance metric is determined based on an updated reward value associated with the updated action, and wherein the updated reward value is larger than the first reward value; and

transmitting the final action for each node to the network of nodes, whereby each node performs the final action in response to the one or more interruptive events.

13. The non-transitory computer-readable medium of claim 12 , wherein setting the one or more interruptive events comprises:

determining a probability distribution of the one or more interruptive events.

14. The non-transitory computer-readable medium of claim 12 , wherein determining an action for each node in response to the one or more interruptive events comprises:

determining the action that are included in the action spaces of the node and in accordance with the attributes and the policies of the node.

15. The non-transitory computer-readable medium of claim 12 , wherein a performance metric of each node comprises whether the node resolves the one or more interruptive events.

16. The non-transitory computer-readable medium of claim 15 , wherein whether the node resolves the one or more interruptive events are determined based on success criteria that include whether a performance threshold is satisfied under a set of predetermined conditions.

17. A system comprising one or more computers comprising one or more processors and one or more storage devices on which are stored instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:

receiving parameter data from a network of nodes, the parameter data comprising attributes, policies, and action spaces for each node included in the network of nodes;

configuring one or more interruptive events on one or more nodes included in the network of nodes;

determining a first action of each node included in the network of nodes in response to the one or more interruptive events;

determining a first performance metric, for each node, that corresponds to the first action, wherein the first performance metric is determined based on at least a first reward value associated with the first action;

performing a reinforcement learning process to train a machine learning model to generate actions to be performed by nodes that improve anti-fragility of the network, including continuously updating the first action for each node in the reinforcement learning process until an anti-fragility metric of a final action satisfies a performance threshold representing the anti-fragility of the network, wherein updating the first action for a node in the reinforcement learning process comprises:

determining an updated action for the node based on the first performance metric of the first action, and

determining an updated performance metric, for the node, that corresponds to the updated action, wherein the updated performance metric is determined based on an updated reward value associated with the updated action, and wherein the updated reward value is larger than the first reward value; and

transmitting the final action for each node to the network of nodes, whereby each node performs the final action in response to the one or more interruptive events.

18. The system of claim 17 , wherein setting the one or more interruptive events comprises:

determining a probability distribution of the one or more interruptive events.

19. The system of claim 17 , wherein determining an action for each node in response to the one or more interruptive events comprises:

determining the action that are included in the action spaces of the node and in accordance with the attributes and the policies of the node.

20. The system of claim 17 , wherein a performance metric of each node comprises whether the node resolves the one or more interruptive events.