Pattern Recognition Parameter Tuning via Segment Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pattern recognition methods, such as those using Hidden Markov Models (HMMs) for tasks like automatic speech recognition, face challenges in effectively distinguishing correct patterns from numerous erroneous possibilities during discriminative training, particularly in optimizing classification errors and mutual information.
Innovation Solution
The approach involves updating pattern model parameters by generating training patterns where one segment of a reference pattern is replaced with a corresponding competing segment, allowing for fine-tuning of classification parameters one segment at a time, using methods like Minimum Classification Error (MCE) and Maximum Mutual Information (MMI) to increase the correspondence of the reference pattern with the pattern signal while decreasing correspondence with competing patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pattern model parameters are updated using competing patterns that are identical to the reference pattern except for one segment at a time, then the precision of classification parameters is improved, but the complexity of training process increases
Solution Approach 1:
The patent applies segmentation by dividing the pattern into segments and replacing one segment at a time with competing segments during training. This allows the system to focus on distinguishing characteristics of individual segments rather than the entire pattern, thereby improving parameter precision while managing training complexity through systematic decomposition of the learning task.
2Reliability
If the HMM parameters are trained by optimizing an objective function related to classification errors, then the ability to discriminate between correct and incorrect patterns is improved, but the difficulty of modeling distinguishing characteristics increases
Solution Approach 1:
The patent applies local quality by focusing the training process on individual segments and their distinguishing characteristics rather than treating the entire pattern uniformly. By replacing one segment at a time with competing segments and adjusting parameters locally, the system effectively models the distinguishing characteristics of specific segments, improving the overall discriminative ability while reducing the complexity of modeling.
3Ease of operation
If pattern segments are replaced one at a time to generate training patterns, then the ease of parameter tuning is improved, but the time required for training increases
Solution Approach 1:
The patent segments the pattern into smaller units and performs parameter tuning by replacing one segment at a time. This segmentation approach makes the tuning process more manageable and systematic, improving ease of operation. While it may increase training time compared to holistic approaches, the structured nature of segment-by-segment replacement provides clearer control over the training process and improves parameter precision.
Data Source
AI summary
Pattern model parameters are updated using update equations based on competing patterns that are identical to a reference pattern except for one segment at a time that is replaced with a competing segment. This allows pattern recognition parameters to be tuned one segment at a time, rather than have to try to model distinguishing features of the correct pattern model as a whole, according to an illustrative embodiment. A reference pattern and competing patterns are divided into pattern segments. A set of training patterns are generated by replacing one of the pattern segments in the reference pattern with a corresponding competing pattern segment. For each of the training patterns, a pattern recognition model is applied to evaluate a relative degree of correspondence of the reference pattern with the pattern signal compared to a degree of correspondence of the training patterns with the pattern signal.