Numerical-to-Spike Conversion with Randomized Spike-Train Construction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current spiking neural networks face issues with slow spike generation speed and inefficient power consumption due to high complexity and unfriendly hardware implementation, particularly when using Poisson distributions.
Innovation Solution
A method and apparatus for converting numerical values into spikes that generate a random spike sequence with improved compatibility to biological characteristics and reduced complexity, involving the generation of consecutive spikes followed by random data modification to form a target spike sequence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If Poisson distribution is used to generate random spike sequence, then biological compatibility is improved, but spike generation speed and power efficiency deteriorate
Solution Approach 1:
The patent segments the Poisson distribution generation process into two independent parts: (1) determining the number of spikes using a simplified random mechanism, and (2) distributing these spikes across time bins using uniform random selection. This segmentation avoids the computational complexity of direct Poisson distribution while maintaining its statistical properties, thereby improving spike generation speed without sacrificing biological compatibility.
Solution Approach 2:
The patent changes the parameters and methodology of random number generation. Instead of using complex Poisson distribution calculations, it uses simple uniform random number generation to determine spike counts and positions. This parameter change maintains the statistical characteristics of Poisson-distributed spike trains while dramatically reducing computational complexity and improving generation speed.
2Adaptability or versatility
If Poisson distribution is used to generate random spike sequence, then biological compatibility is improved, but device complexity and hardware implementation difficulty increase
Solution Approach 1:
The patent replaces complex, resource-intensive Poisson distribution computation with simple, lightweight random number generation mechanisms. This substitution uses computationally inexpensive operations (uniform random sampling) to achieve the same functional outcome, thereby reducing hardware complexity and implementation difficulty while preserving biological compatibility.
Solution Approach 2:
The patent fundamentally changes the mathematical parameters and algorithms used for spike generation. By replacing Poisson distribution calculations with uniform random sampling for spike count determination and position selection, it achieves the same biological realism with significantly simpler hardware requirements and lower implementation complexity.
3Loss of information
If complex distribution methods are used for spike generation, then spike sequence randomness is improved, but power consumption efficiency deteriorates
Solution Approach 1:
The patent extracts only the essential randomness requirement from the Poisson distribution model, separating it from the computationally expensive calculation components. By using simple uniform random number generation to achieve the necessary randomness in spike timing and count, it eliminates the energy-consuming complex calculations while preserving the critical randomness property needed for biological fidelity.
Solution Approach 2:
The patent changes the computational parameters from complex Poisson distribution calculations to simple uniform random sampling. This parameter transformation maintains the randomness and unpredictability of spike sequences (critical for biological compatibility) while dramatically reducing the computational power required, thus improving power consumption efficiency.
Data Source
Figure 1~3
Figure 4
Figure 5~8
AI summary
A method for converting numerical values to spikes. The method comprises: generating an initial spike sequence according to input numerical values (S110), wherein the initial spike sequence comprises at least one data string, each of the at least one data string is independently selected from one of a continuous spike train or a continuous spikeless train, the number of spikes in all of the at least one data string is equal to an expected value of the number of spikes in a target spike sequence to be generated, and the target spike sequence is a spike sequence of a spiking neural network in a time period; and randomly selecting data from the initial spike sequence and changing same, so as to form the target spike sequence (S120). The method enables generation of a random spike sequence having good conformity with biological properties, and having low costs. Further provided are an apparatus for converting numerical values to spikes, an electronic device, and a computer-readable storage medium.