Neural Network Anomaly Localization in Spatio-Temporal Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for locating anomalies in spatio-temporal data face challenges such as the difficulty of obtaining annotated data sets, inability to learn during the production phase, and inability to adapt to new environments or detect new types of anomalies.
Innovation Solution
The method employs reinforcement learning and similarity learning of neural networks during the production phase, allowing the network to learn from unannotated spatio-temporal data, adapt to new environments, and detect new anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning methods with pre-trained neural networks are used, then the system can locate anomalies in spatio-temporal data, but the system cannot learn or adapt during the production phase
Solution Approach 1:
The patent implements a feedback mechanism where user corrections of anomaly locations are fed back into the neural network during the production phase. This allows the system to continuously learn and improve its anomaly detection capabilities without requiring complete retraining, resolving the contradiction between maintaining reliability and enabling adaptability.
Solution Approach 2:
The system transitions from a static pre-trained model to a dynamic model that can adapt its parameters during production based on user feedback. This dynamic adjustment enables the neural network to evolve and improve its performance over time while maintaining its core anomaly detection functionality.
2Reliability
If annotated spatio-temporal data sets are used for training, then the neural network can be trained to locate anomalies, but obtaining such annotated data sets is complex, long and costly
Solution Approach 1:
The system enables users to directly correct and annotate anomaly locations during production use, transforming end-users into active participants in the data annotation process. This eliminates the need for separate, expensive annotation phases and allows the system to improve continuously through actual usage.
Solution Approach 2:
The patent performs preliminary anomaly detection automatically, presenting pre-processed results to users for correction. This preliminary action reduces the annotation burden on users, as they only need to correct errors rather than annotate from scratch, significantly reducing the time and cost investment required.
3Measurement precision
If a neural network is trained on predetermined anomalies, then it can locate those specific anomalies, but it cannot identify new types of anomalies or adapt to new environments
Solution Approach 1:
User feedback on corrected anomaly locations provides the neural network with information about new anomaly types and patterns encountered in practice. This continuous feedback loop enables the system to expand its detection capabilities beyond the predetermined anomalies in its training set.
Solution Approach 2:
The system allows the neural network to adjust its internal parameters and decision boundaries based on user corrections, enabling it to adapt to new anomaly types by changing its detection parameters rather than requiring complete retraining on new data sets.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
One aspect of the invention relates to a method (100) for locating at least one anomaly in a spatio-temporal data item. The method comprises in particular steps of: - Obtaining (130) a spatio-temporal data item, - Obtaining (140) a neural network configured to generate location information for at least one anomaly from a spatio-temporal data item, - Generating (150), by the obtained neural network, the location information for the at least one anomaly by providing the neural network with the spatio-temporal data item obtained, - Obtaining (160) an accuracy score provided by a user evaluating an accuracy of the generated location information for the at least one anomaly, - Reinforcement learning (170) of the neural network, and - Similarity learning (180) of the neural network learned by reinforcement (170).