Stochastic Hyperdimensional Arithmetic Computing for Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hyperdimensional computing (HDC) solutions are weak in feature extraction from complex data such as image and video data, requiring expensive pre-processing algorithms.
Innovation Solution
The StocHD system introduces stochastic hyperdimensional arithmetic computing, enabling end-to-end hyperdimensional learning over raw data by mathematically defining stochastic arithmetic over HDC hypervectors and utilizing a novel fully digital and scalable processing in-memory (PIM) architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing HDC solutions use pre-processing algorithms for feature extraction, then feature extraction capability is improved, but computational complexity and energy consumption increase
Solution Approach 1:
The patent merges the feature extraction function into the HDC learning model by defining stochastic arithmetic operations that work directly on raw data represented as hypervectors. This eliminates the need for separate pre-processing algorithms, combining what were previously distinct processing stages into a unified HDC-based feature extraction and learning framework.
Solution Approach 2:
The stochastic arithmetic operations defined in the patent enable HDC to perform both feature extraction and learning tasks using the same data representation and computational primitives. The binding operation, in particular, serves as a universal mechanism that can extract features from raw data while simultaneously preparing it for learning, making the system multi-functional without requiring separate specialized algorithms.
2Measurement precision
If existing HDC solutions use pre-processing algorithms, then feature extraction is improved, but processing speed decreases
Solution Approach 1:
By merging feature extraction and learning into a single HDC-based process that operates on raw data, the patent eliminates the sequential processing bottleneck where pre-processing must complete before learning can begin. The stochastic arithmetic operations enable both functions to be performed in an integrated manner, improving overall processing throughput.
Solution Approach 2:
The patent enables continuous useful action by allowing the HDC system to process raw data through feature extraction and learning in an uninterrupted flow. The stochastic arithmetic operations maintain data in hypervector representation throughout the entire processing pipeline, eliminating the need to convert between different data representations and enabling continuous processing without idle transitions.
3Measurement precision
If existing HDC solutions use pre-processing algorithms, then feature extraction is improved, but energy consumption increases
Solution Approach 1:
The patent merges multiple processing functions into a single energy-efficient HDC pipeline. By performing feature extraction and learning using the same stochastic arithmetic operations on hypervector representations, the system eliminates redundant computational steps and data conversions that would otherwise consume additional energy in separate pre-processing and learning stages.
Solution Approach 2:
The continuous hypervector representation throughout the processing pipeline enables uninterrupted computation without energy-intensive data format conversions. The stochastic arithmetic operations maintain data in the optimal representation format from input to output, eliminating energy-wasting transitions between different data representations that occur in traditional pre-processing approaches.
Data Source
AI summary
Stochastic hyperdimensional arithmetic computing is provided. Hyperdimensional computing (HDC) is a neurally-inspired computation model working based on the observation that the human brain operates on high-dimensional representations of data, called hypervectors. Although HDC is powerful in reasoning and association of the abstract information, it is weak on feature extraction from complex data. Consequently, most existing HDC solutions rely on expensive pre-processing algorithms for feature extraction. This disclosure proposes StocHD, a novel end-to-end hyperdimensional system that supports accurate, efficient, and robust learning over raw data. StocHD expands HDC functionality to the computing area by mathematically defining stochastic arithmetic over HDC hypervectors. StocHD enables an entire learning application (including feature extractor) to process using HDC data representation, enabling uniform, efficient, robust, and highly parallel computation. This disclosure further provides a novel fully digital and scalable processing in-memory (PIM) architecture that exploits the HDC memory-centric nature to support extensively parallel computation.