Neural Network Reliability Output for Time Series Pattern Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting time series patterns, such as audio patterns, face challenges in reliability and specificity, particularly in speaker authentication tasks, where reducing input dimensionality can lead to misidentifications and failure to capture unique speaker characteristics.
Innovation Solution
The development of artificial neural networks (ANNs) specifically configured to produce both decision and reliability outputs, using neuroevolution of augmenting topologies (NEAT) to build minimal topologies that can weight and ignore outputs based on reliability, allowing for more accurate detection of time series patterns without relying on pre-extracted features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing methods for detecting time series patterns are used, then detection can be performed, but reliability and specificity are reduced leading to misidentifications
Solution Approach 1:
The patent segments the detection task into multiple specialized neural networks, each trained to detect specific time series patterns. This segmentation allows each network to focus on particular patterns, improving both reliability and precision of detection while reducing misidentifications that occur in general-purpose detection methods.
Solution Approach 2:
The patent changes the parameters of the detection system by using neural networks with adjustable weights and thresholds that are optimized for specific pattern types. This parameter optimization enables the system to achieve high reliability and precision by tuning detection sensitivity for different pattern characteristics rather than using fixed detection parameters.
2Device complexity
If input dimensionality is reduced, then processing complexity decreases, but unique speaker characteristics are lost leading to misidentifications
Solution Approach 1:
The patent extracts only the essential features needed for pattern detection directly from raw time series data, avoiding unnecessary dimensionality reduction. The neural networks are designed to process raw or minimally processed input, extracting relevant characteristics while preserving unique speaker properties, thus maintaining reliability without excessive processing complexity.
Solution Approach 2:
The patent transforms the detection approach by adding temporal and contextual dimensions to the analysis. Instead of reducing dimensions, the neural networks analyze time series data in its full temporal context, using sequence modeling to capture speaker characteristics across time, thereby maintaining reliability while managing complexity through efficient temporal processing.
3Adaptability or versatility
If general pattern detection methods are used, then multiple patterns can be detected, but specificity is reduced leading to failure to capture unique characteristics
Solution Approach 1:
The patent creates a universal framework of neural networks that can detect multiple types of time series patterns while maintaining specificity for each pattern type. Each network is designed with architecture and parameters optimized for its target pattern, allowing the system to handle diverse patterns (speech, audio events, contextual sounds) with high precision through specialized yet integrated detection units.
Data Source
AI summary
According to a first aspect of the present disclosure, a method for facilitating detection of one or more time series patterns is conceived, comprising building one or more artificial neural networks, wherein, for at least one time series pattern to be detected, a specific one of said artificial neural networks is built, the specific one of said artificial neural networks being configured to produce a decision output and a reliability output, wherein the reliability output is indicative of the reliability of the decision output. According to a second aspect of the present disclosure, a corresponding computer program is provided. According to a third aspect of the present disclosure, a corresponding system for facilitating the detection of one or more time series patterns is provided.