Meta-Learning Anomaly Detection Model With Limited Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing meta-learning methods for anomaly detection fail to achieve high performance even with a small amount of training data, requiring significant resources for data preparation.
Innovation Solution
A learning method that involves sampling tasks and subsets of data, generating task vectors using neural networks, nonlinearly transforming feature vectors, calculating anomaly scores, and optimizing network parameters to enhance anomaly detection performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large amount of task-specific training data is used to achieve high performance in anomaly detection, then the detection performance is improved, but the cost and complexity of data preparation increases significantly
Solution Approach 1:
The patent applies universality by creating a meta-learning framework that trains a single anomaly detection model capable of performing across multiple different tasks and domains. The model learns universal anomaly detection capabilities from diverse training data (including non-task-specific data) and adapts to specific tasks through meta-learning, eliminating the need to prepare separate task-specific training data for each domain while maintaining high detection performance across various tasks
2Device complexity
If existing meta-learning methods are used to reduce training data requirements, then data preparation cost is reduced, but sufficient performance cannot be achieved
Solution Approach 1:
The patent applies parameter changes by introducing and optimizing specific hyperparameters including the meta-learning rate (lr_meta), task sampling rate (p_sample), and epoch counts (epochs_inner, epochs_outer). The framework dynamically adjusts these parameters during training to optimize the balance between using limited training data and achieving high detection performance. The parameter optimization enables the model to extract maximum information from limited task-specific data while leveraging universal patterns from diverse training data
3Reliability
If task-specific training data is prepared for each anomaly detection task, then the model can be optimized for that specific task, but the time and resources required for preparation increase
Solution Approach 1:
The patent applies preliminary action by pre-training the anomaly detection model on diverse, non-task-specific data before fine-tuning on task-specific data. This preliminary training establishes a robust foundation of universal anomaly detection capabilities that can be quickly adapted to specific tasks. The meta-learning process then efficiently fine-tunes the model parameters using minimal task-specific data, significantly reducing the time and resources required compared to training from scratch for each task
Data Source
AI summary
A learning method includes: receiving as input a set of data sets {D1, . . . , DT} wherein Dt for a task tin a task set {1, . . . , T} includes feature amount vectors of cases of t; sampling t from the task set, and sampling a first subset from Dt and a second subset from Dt excluding the first subset; generating a task vector representing a property oft corresponding to the first subset by a first neural network; nonlinearly transforming feature amount vectors included in data included in the second subset by a second neural network using the task vector; calculating scores representing degrees of anomaly of the feature amount vectors using the transformed feature amount vectors and a preset center vector; and learning parameters of the first and second neural networks so as to make an index value representing generalized performance of anomaly detection higher using the scores.