Apparatus and methods for gating analog and spiking signals in artificial neural networks

1. A computer-implemented method of synaptic gating in a network, the method being performed by one or more processors configured to execute computer program modules, the method comprising:

generating an output at a node of the network, the output configured to inhibit a gating unit of the network;

based on at least one spiking input signal, at least one continuous input signal and a teaching signal, pausing generation of the output; and

based on the pausing, activating the gating unit, thereby effectuating the synaptic gating;

wherein:

generation of the output is characterized by an active node state configured in accordance with a supervised learning process;

the pausing of the output generation is characterized by an inactive node state configured based on the teaching signal; and

the teaching signal comprises one or more spikes configured to transition the node into the inactive state.

2. The method of claim 1 , wherein the output is encoded using a spiking representation.

3. The method of claim 1 , wherein the output is encoded using a continuous representation.

4. The method of claim 1 , further comprising adjusting a characteristic associated with at least one connection of a plurality of synaptic connections of the network in accordance with a parameterized rule having at least one parameter associated therewith;

wherein the characteristic is adjusted based at least in part on modifying the at least one parameter.

5. The method of claim 4 , wherein:

the at least one spiking input signal is received via a first connection of the plurality of connections;

the at least one continuous input signal is received via a second connection of the plurality of connections; and

the at least one connection comprises one of the first connection and the second connection.

6. The method of claim 5 , further comprising modifying, based at least in part on the reception of the at least one continuous input signal, one other characteristic associated with the second connection according to the parameterized rule;

wherein: the at least one connection comprises the first connection.

7. The method of claim 4 , wherein the parameterized rule comprises a parameterized function of the at least one parameter.

8. The method of claim 4 , wherein:

the parameterized rule comprises a supervised learning rule; and

the modifying the at least one parameter is configured based at least in part on a training signal, the target signal being representative of a target node output.

9. A neural network configured to implement synaptic gating in accordance with at least one parameterized rule, the network comprising:

a plurality of connections configured to facilitate transmission of a plurality of spiking and non-spiking signals; and

a plurality of mixed-mode nodes in operative communication with said plurality of connections, the nodes configured to:

generate a node output configured to inhibit one or more synaptic gates; and

based on the at least one parameterized rule, cease generation of the node output for a duration;

wherein in response to the cessation, the one or more synaptic gates are configured to switch one or more signals transmitted via the plurality of connections.

10. The neural network of claim 9 , wherein:

the at least one parameterized rule comprises a parameterized function of at least one parameter; and

a characteristic of at least one of the plurality of nodes is configured to be adjusted based at least in part on the at least one parameter.

11. The neural network of claim 10 , wherein the at least one parameter comprises an integration time constant.

12. The neural network of claim 10 , wherein the at least one parameter comprises one or more of (i) a resting potential (ii) a refractory period, and/or (iii) a firing threshold.

13. The neural network of claim 10 , wherein the at least one parameter comprises a level of stochasticity associated with the generation of the node output.

14. The neural network of claim 10 , wherein the node output is configured based on a reception of at least one of the plurality of spiking input signals and at least one of the plurality of non-spiking input signals at least one of the plurality of nodes.

15. The neural network of claim 14 , wherein the node output comprises at least one of (i) a spiking output and (ii) a non-spiking output.

16. The neural network of claim 14 , wherein one or more of the at least one of the plurality of spiking input signals and the at least one of the plurality of non-spiking input signals comprises a reinforcement signal associated with the at least one parameterized rule.

17. The neural network of claim 14 , wherein:

the node output comprises a tonic spiking output; and

the at least one of the plurality of spiking input signals or at least one of the plurality of non-spiking input signals comprises a teaching signal configured to cause the node to cease output generation for the duration.

18. The neural network of claim 10 , wherein:

the at least one of the plurality of spiking input signals or at least one of the plurality of non-spiking input signals comprises a teaching signal;

the at least one parameterized rule comprises a supervised learning rule configured based on the at least one parameter and the teaching signal;

wherein the teaching signal is configured to characterize a target output of the node determined to use the supervised learning rule; and

wherein the cessation of the output generation by the node is configured to produce the node output consistent with the target output.

19. The neural network of claim 18 , wherein:

the supervised learning rule comprises a logic configured to conduct a sequence of learning trials, where for a given trial of the sequence of trials, the node state is updated based on a value of the learning parameter determined at a preceding trial of the sequence of trials; and

the cessation of the output generation by the node comprises a modification of the parameter in accordance with the value.

20. The neural network of claim 19 , wherein the node output comprises a tonic output encoded by use of an analog representation.

21. The neural network of claim 19 , wherein the node output comprises a tonic output encoded by use of spiking representation.

22. The neural network of claim 10 , wherein:

the node output comprises a spiking signal; and

generation of an output signal at the node is configured based at least in part on a state modification of a state of the node.

23. The neural network of claim 22 , wherein:

the at least one of the plurality of spiking input signals or at least one of the plurality of non-spiking input signals comprises a reinforcement signal;

the at least one parameterized rule comprises a reinforcement learning rule; and

the state modification is configured based at least in part on the reinforcement signal, the reinforcement signal configured to be representative of a target network output.

24. The neural network of claim 23 , wherein the reinforcement signal is encoded by use of any of (i) spiking signal representation and (ii) analog signal representation.

25. The neural network of claim 22 , wherein:

the at least one parameterized rule comprises an unsupervised learning rule; and

the state modification is configured based at least in part on one or more of the at least one spiking signal, the at least one analog signal, and the output signal.

26. The neural network of claim 22 , wherein the state comprises a level of stochasticity associated with generation of the output signal.

27. The neural network of claim 22 , wherein the state comprises at least one of (i) a node excitability, and/or (ii) a node inhibition.

28. The neural network of claim 22 , wherein the state comprises a node susceptibility.