Pattern Search Apparatus Using Cumulative Probability Hashing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approximate nearest neighbor search methods using hash functions face inefficiencies in search time and high error ratios due to varying pattern densities in buckets, and fail when the input pattern does not exist in the same bucket as the training patterns.
Innovation Solution
A pattern search apparatus that uses a storage unit to store training patterns, a distribution acquisition unit to obtain cumulative probability distributions, a hash function unit to divide the distribution into uniform sections, and a search unit to classify patterns into buckets, enabling efficient search by converting input patterns into hash values and finding the most similar training patterns within those buckets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a hash function divides the existence range at constant intervals, then the search process is simplified, but the number of patterns in each bucket varies according to distribution, causing long search times in high-density buckets and high error ratios in low-density buckets
Solution Approach 1:
The patent changes the parameter of hash function division from constant intervals to variable intervals based on cumulative probability distribution. The hash function divides the range such that each bucket contains an equal number of patterns, transforming the distribution parameter to achieve uniform bucket population and consistent search performance.
2Productivity
If the training pattern does not exist in the bucket containing the input pattern, then the search cannot be made, but using constant interval division causes high error ratios in low-density buckets
Solution Approach 1:
The patent modifies the hash function parameter to use cumulative probability distribution-based division instead of constant interval division. This ensures that each bucket contains a consistent number of patterns, improving both search speed and accuracy by preventing empty or sparse buckets while maintaining efficient bucket access.
3Ease of manufacture
If the number of patterns in each bucket varies, then the hash function is simple to implement, but the average search speed decreases due to long search times in high-density buckets
Solution Approach 1:
The patent changes the hash function parameter from constant interval division to cumulative probability distribution-based division. Although this increases implementation complexity slightly, it achieves uniform bucket population distribution, ensuring consistent search time across all buckets and improving overall average search speed.
Data Source
AI summary
A pattern search apparatus includes a storage unit, a distribution acquisition unit, a hash function unit, a training unit, and a search unit. A cumulative probability distribution of the training pattern on an arbitrary axis is obtained, and hash function each of which divides a probability value are defined based on the cumulative probability distribution.