Neuromorphic spike integrator apparatus

1 . A computer system comprising:

one or more computer processors;

one or more computer readable storage medium; and

program instructions stored on the computer readable storage medium for execution by at least one of the one or more processors to cause the computer system to:

receive multiple input signals at a respective arrival time at a neuromorphic spike integrator apparatus;

weight the received signals with a respective weight value corresponding to the respective arrival time, wherein the respective weight value is determined as a respective output of a modulating function using the respective arrival time as respective input to the modulating function, the modulating function being a time-dependent linear function;

integrate the weighted received signals into a current value of a state of the neuromorphic spike integrator apparatus, wherein the current value is adjusted intermittently via a first adjustment and a second adjustment before a spike output of the neuromorphic spike integrator apparatus occurs, with each of the first and second adjustments occurring in response to receiving a respective one of the weighted received signals, and wherein the first adjustment adjusts the current value to a first amount above a reset value and the second adjustment adjusts the current value to a second amount that is greater than the first amount; and

communicate a spike output based on the current value of the state being greater than or equal to a threshold value.

2 . The computer system of claim 1 , wherein the time-dependent linear function is a time-dependent piecewise linear function.

3 . The computer system of claim 1 , wherein the outputs of the time-dependent linear function decrease with increasing arrival time values.

4 . The computer system of claim 1 , wherein the arrival time is determined with respect to a predefined reference point of time based on one of:

(i) a signal that is received at the neuromorphic spike integrator subsequent to a refractory period;

(ii) a signal that is received at the neuromorphic spike integrator apparatus subsequent to generation of an output signal of the neuromorphic spike integrator apparatus;

(iii) a signal that is received at the neuromorphic spike integrator apparatus subsequent to a predefined value of the state of the neuromorphic spike integrator apparatus;

(iv) a signal that is (a) received at the neuromorphic spike integrator apparatus and (b) has an amplitude higher than the current state of the neuromorphic spike integrator apparatus;

(v) a global synchronization signal of the neuromorphic spike integrator apparatus; or

vi) a signal that is received at the neuromorphic spike integrator apparatus subsequent to an inhibition signal.

5 . The computer system of claim 1 , wherein the computer readable storage medium stores an artificial neural network comprising multiple layers, wherein at least one layer of the multiple layers comprises one or more neuron realizations that include a respective instance of the neuromorphic spike integrator apparatus.

6 . The computer system of claim 1 , wherein the neuromorphic spike integrator apparatus is implemented in a hardware accelerator that comprises crossbar arrays.

7 . A computer-implemented method, the method comprising:

receiving at least one input signal at a respective arrival time at a neuromorphic spike integrator neuron apparatus, wherein the at least one input signal represents a value;

weighting, by one or more processors, the received signal with a weight value corresponding to the arrival time, the weighting comprising a multiplication of the weight value and the value, wherein the weight value is determined as an output of a modulating function using the respective arrival time as respective input to the modulating function;

integrating, by one or more processors, the weighted received signal into a current value of a state of the neuromorphic spike integrator apparatus, wherein the current value is adjusted intermittently, with each intermittent adjustment occurring in response to a respective new input signal received; and

communicating, by one or more processors, a signal based on the current value of the state.

8 . The computer-implemented method of claim 7 ,

wherein the modulating function is a piecewise linear function.

9 . The computer-implemented method of claim 7 ,

wherein the modulating function is a time dependent piecewise linear function.

10 . The computer-implemented method of claim 9 , the method further comprising:

the modulating function having values decreasing with increasing arrival time values.

11 . The computer-implemented method of claim 7 ,

wherein the modulating function is a time dependent nonlinear function involving a predefined range of values of encoded information.

12 . The computer-implemented method of claim 11 , the method further comprising:

the modulating function is defined as: R/2 Δt+1 , where R is the range of value of the encoded information and Δt is indicative of the time arrival of the signal.

13 . The computer-implemented method of claim 11 , wherein

the neuromorphic spike integrator apparatus comprises a shift term for shifting the value of the modulating function at the arrival time of a given signal, and incorporating a decay behavior of a state variable based on the shift term, wherein the shift term comprises a constant term indicative of an expected value of a future coming signal after the given signal.

14 . The computer-implemented method of claim 7 , the method further comprising:

determining, by the one or more processors, the arrival time with respect to a predefined reference point of time, wherein the reference point of time is based on one of:

(i) a signal that is received at the neuromorphic spike integrator apparatus subsequent to a refractory period;

(ii) a signal that is received at the neuromorphic spike integrator apparatus subsequent to generation of an output signal of the neuromorphic spike integrator apparatus;

(iii) a signal that is received at the neuromorphic spike integrator apparatus subsequent to a predefined value of the state of the neuromorphic spike integrator apparatus;

(iv) a signal that is (a) received at the neuromorphic spike integrator apparatus and (b) has an amplitude higher than the current state of the neuromorphic spike integrator apparatus;

(v) a global synchronization signal of the neuromorphic spike integrator apparatus; or

(vi) a signal that is received at the neuromorphic spike integrator apparatus subsequent to an inhibition signal.

15 . A computer program product comprising:

one or more computer-readable storage media and program instructions stored on the one or more computer-readable storage media, the program instructions comprising:

program instructions to receive at least one input signal encoding information at a neuromorphic spike integrator apparatus, the receiving occurring at a respective arrival time;

program instructions to weight the received signal with a weight value, wherein the weight value is determined as a value of a modulating function using the arrival time as input to the modulating function, the modulating function being a time dependent nonlinear curved function involving a predefined range of values of the input signal encoding information;

program instructions to integrate the weighted received signal into a current value of a state of the neuromorphic spike integrator apparatus; and

program instructions to communicate a signal based on the current value of the state.

16 . The computer program product of claim 15 , wherein the modulating function is defined as:

R

2

Δ

⁢

t

+

1

,

where R is the range of value of the encoded information and Δt is indicative of the arrival time of the signal.

17 . The computer program product of claim 16 , the program instructions further comprising:

program instructions to modify the value of the modulating function at the arrival time of a given signal based on a shift term that incorporates a decay behavior of a state variable, wherein the shift term includes a constant term indicative of an expected value of a future coming signal after the given signal.

18 . The computer program product of claim 15 , the program instructions further comprising:

program instructions to determine the arrival time with respect to a predefined reference point of time, wherein the reference point of time is based on one of:

(i) a signal that is received at the neuromorphic spike integrator subsequent to a refractory period;

(ii) a signal that is received at the neuromorphic spike integrator apparatus subsequent to generation of an output signal of the neuromorphic spike integrator apparatus;

(iii) a signal that is received at the neuromorphic spike integrator apparatus subsequent to a predefined value of the state of the neuromorphic spike integrator apparatus;

(iv) a signal that is (a) received at the neuromorphic spike integrator apparatus and (b) has an amplitude higher than the current state of the neuromorphic spike integrator apparatus;

(v) a global synchronization signal of the neuromorphic spike integrator apparatus; or

(vi) a signal that is received at the neuromorphic spike integrator apparatus subsequent to an inhibition signal.

19 . The computer program product of claim 15 , the program instructions further comprising

program instructions to generate an artificial neural network comprising multiple layers, wherein at least one layer of the multiple layers comprises one or more neuron realizations that include the neuromorphic spike integrator apparatus.

20 . The computer program product of claim 15 , the program instructions further comprising:

program instructions to modify the value of the modulating function at the arrival time of a given signal based on a shift term that incorporates a decay behavior of a state variable, wherein the shift term is a time dependent term determined dynamically from spiking statistics of previously received signals.