Monitoring neural networks with shadow networks

1. A method for generating an event, comprising

monitoring an activity of a first neural network with a second neural network; and

generating the event, at the second neural network, based at least in part on a signal received from a first neuron of the first neural network, during the monitoring, the received signal causing an activity of a second neuron of the second neural network to exceed an exception threshold.

2. The method of claim 1 , in which operation of the first neural network is independent from operation of the second neural network.

3. The method of claim 1 , in which the first neural network has same neuron types as the second neural network.

4. The method of claim 1 , in which the first neural network has a same topology as the second neural network.

5. The method of claim 1 , in which the event is an exception.

6. The method of claim 5 , in which the exception is a breakpoint for debugging.

7. The method of claim 1 , in which the event triggers a message.

8. The method of claim 7 , in which the message affects operation for a network, system, and/or hardware associated with the first neural network, the network, the system, and the hardware being different from the first neural network.

9. The method of claim 1 , further comprising reconfiguring the second neural network based at least in part on the event.

10. The method of claim 9 , in which reconfiguring comprises one or more of adjusting a size of the second neural network, reconfiguring parameters, and/or adjusting thresholds.

11. The method of claim 1 , in which the second neural network is further configured to generate the event based at least in part on a subset of states of the first neural network.

12. The method of claim 11 , in which a second state of the second neural network tracks a first state of the first neural network.

13. The method of claim 12 , in which:

the first neuron of the first neural network operates based at least in part on the first state, the first neuron having a first threshold, and

the second neuron of the second neural network operates based at least in part on the second state, the second neuron having the exception threshold.

14. The method of claim 1 , in which the second network is trainable based at least in part on one or more of a user input, monitoring performed by the second network, or a combination thereof.

15. An apparatus for generating an event, the apparatus comprising:

a memory unit; and

at least one processor coupled to the memory unit; the at least one processor being configured:

to monitor an activity of a first neural network with a second neural network; and

to generate the event, at the second neural network, based at least in part on a signal received from a first neuron of the first neural network, during the monitoring, the received signal causing an activity of a second neuron of the second neural network to exceed an exception threshold.

16. The apparatus of claim 15 , in which operation of the first neural network is independent from operation of the second neural network.

17. The apparatus of claim 15 , in which the first neural network has same neuron types as the second neural network.

18. The apparatus of claim 15 , in which the first neural network has a same topology as the second neural network.

19. The apparatus of claim 15 , in which the event is an exception.

20. The apparatus of claim 19 , in which the exception is a breakpoint for debugging.

21. The apparatus of claim 15 , in which the event triggers a message.

22. The apparatus of claim 21 , in which the message affects operation for a network, system, and/or hardware associated with the first neural network, the network, the system, and the hardware being different from the first neural network.

23. The apparatus of claim 15 , in which the at least one processor is further configured to reconfigure the second neural network based at least in part on the event.

24. The apparatus of claim 23 , in which the at least one processor is further configured to perform one or more of adjusting a size of the second neural network, reconfiguring parameters, adjusting thresholds, or a combination thereof based at least in part on the event.

25. The apparatus of claim 15 , in which the second neural network is further configured to generate the event based at least in part on a subset of states of the first neural network.

26. The apparatus of claim 25 , in which a second state of the second neural network tracks a first state of the first neural network.

27. The apparatus of claim 26 , in which:

the first neuron of the first neural network operates based at least in part on the first state, the first neuron having a first threshold, and

the second neuron of the second neural network operates based at least in part on the second state, the second neuron having the exception threshold.

28. The apparatus of claim 15 , in which the second neural network is trainable based at least in part on one or more of a user input, monitoring performed by the second network, or a combination thereof.

29. An apparatus for generating an event, the apparatus comprising:

means for monitoring an activity of a first neural network with a second neural network; and

means for generating the event, at the second neural network, based at least in part on a signal received from a first neuron of the first neural network, during the monitoring, the received signal causing an activity of a second neuron of the second neural network to exceed an exception threshold.

30. A non-transitory computer-readable medium having program code recorded thereon for generating an event, the program code comprising:

program code to monitor an activity of a first neural network with a second neural network; and

program code to generate the event, at the second neural network, based at least in part on a signal received from a first neuron of the first neural network, during the monitoring, the received signal causing an activity of a second neuron of the second neural network to exceed an exception threshold.