LOES Encoding for Compact Pattern Databases in Best-First Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory-based search heuristics in domain-independent planning face challenges with memory efficiency, particularly when using best-first search algorithms with duplicate detection, as they often require excessive memory and are not well-suited for complex subproblems, leading to inefficiencies in pattern database representations.
Innovation Solution
The implementation of a Level-Ordered Edge Sequence (LOES) encoding scheme for prefix trees, which allows for a compact representation of pattern databases, enabling efficient navigation and storage by using a two-level index and omitting complete subtrees through a special code-point, thereby reducing memory requirements and improving processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If best-first search algorithms with duplicate detection are used, then solution optimality is improved, but memory requirements increase excessively
Solution Approach 1:
The patent segments the pattern database into multiple smaller databases, each covering a specific subset of patterns. This segmentation reduces the memory footprint of each individual database while maintaining the ability to provide optimal heuristic guidance through coordinated access to multiple segments.
Solution Approach 2:
The patent introduces a new dimension of organization by using a hierarchical structure with a root database and multiple child databases. This dimensional change allows the system to manage large pattern spaces efficiently by distributing patterns across multiple levels and subsets, reducing peak memory requirements.
2Productivity
If pattern database size is increased to improve heuristic accuracy, then search performance is improved, but memory consumption increases
Solution Approach 1:
The patent divides large pattern databases into smaller segmented databases, each optimized for specific pattern subsets. This allows the system to achieve high search performance through coordinated access to multiple specialized databases while keeping individual memory consumption manageable.
Solution Approach 2:
Each segmented database is optimized with local quality tailored to its specific pattern subset, using customized encoding schemes and data structures. This local optimization ensures high performance for relevant patterns while minimizing memory usage for each segment.
3Speed
If perfect hashing is used for subproblem representation, then access efficiency is improved, but adaptability to complex subproblems deteriorates
Solution Approach 1:
The patent employs dynamic hashing strategies that adapt to the complexity and characteristics of different subproblems. The hashing function and database structure can be adjusted based on the specific pattern subset being processed, maintaining both access efficiency and adaptability to complex scenarios.
Solution Approach 2:
The system changes hashing parameters and data organization based on the specific subproblem characteristics. Different hashing functions, bucket sizes, and organization schemes are applied to different pattern subsets, optimizing both access efficiency and adaptability for each case.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Compressed Level-Ordered Edge Sequence (CLOES) encodings are described. These techniques enable more aggressive compression of a state-set representation.