Encoding Circuit for Hyperdimensional Computing Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reducing the cost of feature storing and feature searching in In-Memory Searching (IMS) memory, such as hyperdimensional computing, have limited capacity and efficiency, necessitating improvements in encoding techniques for AI feature vector learning.
Innovation Solution
An encoding method and circuit that perform linear conversion using a convolution layer, followed by activation, binding with random vectors, and processing with a Signum function and normalization to generate an output vector, enhancing the efficiency of feature encoding and storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If hyperdimensional computing methodology is used for binary quantization, then encoding efficiency is improved, but encoding capacity is limited
Solution Approach 1:
The encoding process is divided into multiple stages: convolution layer for linear conversion, activation function for binary quantization, binding circuit for random vector combination, adding circuit for aggregation, and Signum normalization for final output. This segmentation allows each stage to optimize for specific functions, achieving both efficiency and capacity
Solution Approach 2:
The patent transitions from traditional hyperdimensional computing to a multi-dimensional processing approach using convolution layers and random vector binding. This dimensional transformation enables the system to achieve both high encoding efficiency and expanded capacity by operating in multiple feature spaces simultaneously
2Quantity of substance
If 32-bit floating point features are quantized into binary features, then feature storing and searching cost is reduced, but quantization accuracy is compromised
Solution Approach 1:
The patent applies parameter changes through the activation function that compares convolution output with reference values, and the Signum normalization that transforms adding results into final binary outputs. These parameter transformations enable accurate quantization while maintaining binary format benefits for storage and searching
Solution Approach 2:
The activation function and Signum normalization act as feedback mechanisms that adjust the quantization process. The activation function compares intermediate results with reference values, and the Signum function processes adding results to generate final outputs, ensuring accurate quantization throughout the encoding pipeline
Data Source
AI summary
The application provides an encoding method and an encoding circuit. The encoding method includes: performing linear conversion on an input into a first vector based on a weight by a convolution layer; comparing the first vector generated from the convolution layer with a reference value to generate a second vector by an activation function; binding the second generated by the activation function with a random vector to generate a plurality of binding results; adding the binding results to generate an adding result; and operating the adding result by a Signum function and a normalization function to generate an output vector.