Inference Apparatus Non-Zero Element Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in reducing the number of multiplication operations in the convolutional layer of deep learning, particularly in image recognition using multilayer neural networks, as they require extensive computational resources.
Innovation Solution
An inference apparatus utilizing multiple Processing Elements (PEs) that execute convolution operations by identifying and multiplying only non-zero elements of input data with corresponding weights, thereby reducing the number of multiplication operations in the convolutional layer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional convolution operations are performed in the convolutional layer, then complete feature extraction is achieved, but the number of multiplication operations becomes excessively large
Solution Approach 1:
The patent extracts and identifies non-zero elements from the input data in the convolutional layer, then performs multiplication operations only on these extracted non-zero elements with their corresponding weights. This selective extraction approach maintains complete feature extraction accuracy while significantly reducing the number of multiplication operations compared to conventional full-element convolution.
2Use of energy by moving object
If the number of multiplication operations is reduced by skipping zero elements, then computational load decreases, but processing complexity increases due to element identification
Solution Approach 1:
The patent performs preliminary identification of non-zero elements and their positions before the multiplication operation phase. By pre-processing the input data to mark or store non-zero element locations, the system avoids repeated zero-checking during multiplication, thereby reducing overall computational load while managing processing complexity through structured preliminary action.
Data Source
AI summary
An inference apparatus comprises a plurality of PEs (Processing Elements) and a control part. The control part operates a convolution operation in a convolutional neural network using each of a plurality of pieces of input data and a weight group including a plurality of weights corresponding to each of the plurality of pieces of input data by controlling the plurality of PEs. Further, each of the plurality of PEs executes a computation including multiplication of a single piece of the input data by a single weight and also executes multiplication included in the convolution operation using an element with a non-zero value included in each of the plurality of pieces of input data.