Sparse Neuromorphic Processor Parallel Convolutional Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep learning algorithms require intense computational resources and high power consumption due to their inability to effectively utilize sparsity, leading to inefficiencies in processing and classification tasks.
Innovation Solution
A processor architecture that leverages sparsity by using inference and classification modules with parallel convolutional operations to generate sparse representations, reducing computational needs and power consumption through optimized parallel processing and sparse convolvers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deep learning algorithms use multiple layers of specialized kernels for detection and classification, then classification accuracy is improved, but computational resources and power consumption increase intensely
Solution Approach 1:
The patent changes the parameter of data representation from dense to sparse format. By representing data sparsely (where most values are zero or negligible), the system maintains classification accuracy while dramatically reducing the number of computations required, thereby lowering power consumption in neuromorphic processing systems.
Solution Approach 2:
The patent extracts and processes only the significant non-zero elements from sparse data representations. By taking out and processing only the relevant portions of data rather than all elements, the system reduces computational workload and power consumption while maintaining classification accuracy.
2Measurement precision
If the depth of the neural network increases to improve feature extraction, then classification performance is improved, but computational intensity and memory size increase significantly
Solution Approach 1:
The patent applies sparse representation parameters to neural network data, allowing deeper networks to be implemented with reduced computational intensity. By changing the data format to sparse, the system can maintain complex deep architectures without proportionally increasing computational burden.
Solution Approach 2:
The patent implements partial processing by focusing computational resources only on non-zero elements of sparse data. This partial action approach allows deep networks to process only necessary data portions, reducing overall computational intensity while maintaining feature extraction capability.
3Reliability
If conventional processors process all data elements uniformly, then completeness of processing is maintained, but sparsity potential is wasted leading to inefficiency
Solution Approach 1:
The patent applies different processing qualities to different data elements based on their significance. Non-zero elements receive full processing attention while zero or negligible elements are skipped or handled minimally. This local quality differentiation maintains processing completeness for important data while improving overall efficiency through selective processing.
Data Source
AI summary
An information processor is provided that includes an inference module configured to extract a subset of data from information in an input and a classification module configured to classify the information in the input based on the extracted data. The inference module includes a first plurality of convolvers acting in parallel to apply each of N1 convolution kernels to each of N2 portions of the input image in order to generate an interim sparse representation of the input and a second plurality of convolvers acting in parallel to apply each of N3 convolution kernels to each of N4 portions of the interim sparse representation to generate a final sparse representation containing the extracted data. In order to take advantage of sparsity in the interim sparse representation, N3 is greater than N4 to parallelize processing in a non-sparse dimension and/or the second plurality of convolvers comprise sparse convolvers.