Multi-Modal Anomaly Detection Using Joint Anomaly Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for anomaly detection in multi-modal data streams struggle with the complexity of systems where labeled data is rare or nonexistent, leading to inefficiencies in identifying anomalies such as defective products, viruses, threats, and disease diagnoses, and require assumptions about correlated data streams.
Innovation Solution
A method that trains models using joint anomalies and thresholds to identify anomalies in multi-modal data streams, updating parameters and confidence levels to enhance detection accuracy without relying on pre-labeled datasets, utilizing covariance maximization and unsupervised metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional anomaly detection methods are used that assume labeled training data is available, then the detection process can be simplified, but this approach fails when labeled data is rare or nonexistent as stated in the background
Solution Approach 1:
The patent inverts the traditional anomaly detection approach by assuming anomalies are rare and labeled data is unavailable, then using unlabeled data to train models through self-supervised learning and anomaly scoring mechanisms rather than relying on pre-labeled training sets
Solution Approach 2:
The system performs self-service by using the available unlabeled multi-modal data to train and improve its own anomaly detection models through self-supervised learning, eliminating the need for external labeled data sources
2Measurement precision
If multiple data streams from different sensors are analyzed together to improve detection accuracy, then the precision of anomaly identification increases, but the complexity of the system increases
Solution Approach 1:
The patent segments the multi-modal data processing by training separate anomaly detection models for each data stream or sensor type, then combining their outputs through ensemble methods or anomaly scoring, rather than processing all data streams as a single complex unit
Solution Approach 2:
The system uses universal anomaly detection models that can process multiple types of data streams through a unified framework, with models trained to handle various sensor types and data formats through self-supervised learning
3Reliability
If the threshold for anomaly detection is lowered to increase sensitivity, then more anomalies are detected, but the number of false positives increases
Solution Approach 1:
The system implements feedback mechanisms where anomaly detection results are continuously evaluated and used to adjust detection thresholds and model parameters, allowing the system to learn from false positives and improve its precision over time through self-supervised learning
Data Source
AI summary
Systems and method for detecting anomalies in accordance with embodiments of the invention are illustrated. One embodiment includes receiving a first data stream collected from a first sensor, identifying a first set of anomalies in the first data stream using a first model, receiving a second data stream collected from a second sensor, identifying a second set of anomalies in the second data stream using a second model, determining a set of joint anomalies using the first set of anomalies, second set of anomalies, and a threshold, wherein a threshold is some time period, and updating how anomalies are identified in the first and second set of models using the set of joint anomalies.