Hyperdimensional Computing for Zero-Shot Image Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current zero-shot learning methods, particularly those using generative adversarial networks, face inefficiencies in generating constructed images, as they rely on unsupervised learning and are time-consuming.
Innovation Solution
The proposed solution employs hyperdimensional computing (HDC) to generate and integrate image and semantic hyperdimensional vectors, allowing for rapid construction of images and inference by using neural networks for feature extraction and content-addressable memory for high-speed matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If generative adversarial networks are used for zero-shot learning, then recognition accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent replaces the mechanical training process of generative adversarial networks with a hyperdimensional computing system that uses algebraic operations on high-dimensional vectors. Instead of iterative gradient descent and multiple training epochs, the system represents images and semantics as hyperdimensional vectors and performs rapid matching using hyperdimensional operations such as binding, mixing, and matching, achieving both accuracy and speed.
Solution Approach 2:
The patent transforms the problem from traditional image space to hyperdimensional space by changing the dimensional parameters. Images are encoded into hyperdimensional vectors with dimensions much larger than the original pixel space, allowing for more discriminative representation. This parameter transformation enables rapid comparison and matching operations that are computationally efficient while maintaining high recognition accuracy.
2Productivity
If traditional image processing methods are used, then processing speed is maintained, but recognition accuracy for unseen objects deteriorates
Solution Approach 1:
The patent moves from traditional 2D or 3D image space to hyperdimensional space with thousands or millions of dimensions. This dimensional expansion allows the system to capture complex semantic relationships and subtle image features that are invisible in lower dimensions. The hyperdimensional representation enables the system to recognize unseen objects by matching semantic descriptions with hyperdimensional image vectors, achieving high accuracy while maintaining fast processing through algebraic operations.
Solution Approach 2:
The patent creates a unified hyperdimensional framework that simultaneously handles image representation, semantic encoding, and recognition matching. The same hyperdimensional vectors and operations serve multiple functions: representing images, encoding semantic descriptions, performing feature extraction, and enabling rapid comparison. This multi-functionality eliminates the need for separate processing pipelines, maintaining processing speed while improving recognition accuracy across diverse tasks.
3Measurement precision
If more training data is collected, then recognition accuracy improves, but system complexity and resource requirements increase
Solution Approach 1:
The patent uses hyperdimensional binding operations to create compact representations that capture essential features and relationships. Instead of storing and processing large amounts of raw training data, the system creates hyperdimensional vector copies that encode semantic and visual information in a compressed form. These hyperdimensional representations can be rapidly compared and matched without requiring the original large datasets, reducing system complexity while maintaining recognition accuracy.
Data Source
AI summary
A non-transitory computer-readable recording medium stores a program for causing a computer to execute a process including for each of plural pieces of first-type training data including image information, first semantic information, and a first class of a relevant first object, generating a first hyperdimensional vector (HV) from the image information and the first semantic information, and storing the first HV in a storage unit in correlation with the first class, and for each of plural pieces of second-type training data including second semantic information and a second class of a relevant second object, obtaining, from the storage unit, a predetermined number of HVs exhibiting a higher degree of matching with an HV generated from the second semantic information, generating a second HV of the second-type training data based on the predetermined number of HVs, and storing the second HV in the storage unit in correlation with the second class.