Weighted Finite State Automaton Deformation via Preliminary Weight Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for deforming weighted finite state automata (WFSA) do not effectively eliminate unnecessary paths generated during weight quantization, leading to inefficient memory usage and suboptimal structure for pattern recognition applications.
Innovation Solution
A device and method that transforms weights prior to deformation to reduce the number of elements and preserve order, allowing for a smaller number of transitions and reduced memory requirements by implementing weight transformation operations such as rounding, bit reduction, and constant multiplication, followed by conventional deformation operations like determinization and minimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If weight quantization is performed after deformation operation, then memory size is reduced, but unnecessary paths cannot be eliminated
Solution Approach 1:
The patent applies preliminary action by performing weight transformation (quantization) before the deformation operation instead of after. This preliminary weight transformation reduces the number of distinct weight values, which subsequently enables the deformation operation to eliminate more unnecessary paths and produce a more compact automaton structure with fewer transitions.
Solution Approach 2:
The patent changes the weight parameters of the WFSA through transformation operations (such as rounding to integers or reducing precision) before performing deformation. This parameter change reduces the complexity of the weight space, allowing the deformation operation to more effectively merge states and eliminate unnecessary paths, thereby reducing both memory size and structural complexity.
2Device complexity
If conventional deformation operation is performed without prior weight transformation, then structure is simplified, but memory usage is not optimized
Solution Approach 1:
The patent applies preliminary action by performing weight transformation (quantization) before the deformation operation instead of after. This preliminary weight transformation reduces the number of distinct weight values, which subsequently enables the deformation operation to more effectively merge states and eliminate unnecessary paths, thereby producing a more compact automaton structure with fewer transitions.
Solution Approach 2:
The patent changes the weight parameters of the WFSA through transformation operations (such as rounding to integers or reducing precision) before performing deformation. This parameter change reduces the complexity of the weight space, allowing the deformation operation to more effectively merge states and eliminate unnecessary paths, thereby reducing both memory size and structural complexity.
Data Source
AI summary
According to an embodiment, an automaton deforming device includes a transforming unit and a deforming unit. The transforming unit generates second values by transforming first values, which either represent weights assigned to transitions in a weighted finite state automaton or represent values that are transformed into weights assigned to transitions in a weighted finite state automaton, in such a way that number of elements of a set of the first values are reduced and an order of the first values is preserved. The deforming unit deforms a weighted finite state automaton in which weights according to the second values are assigned to transitions.