Edge Anomaly Detection Using Quantized Random Tree Ensembles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Anomaly detection consumes significant compute and memory resources, making it challenging to implement on resource-constrained devices, particularly due to the lack of labeled datasets and varied types of anomalies.
Innovation Solution
An unsupervised anomaly detection process using optimal data quantization and ensemble of random decision trees, which reduces training samples and generates a more efficient model by partitioning data into bins and selecting split points based on data density, facilitating anomaly detection with reduced power and resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If anomaly detection is implemented using traditional methods, then detection accuracy is improved, but compute and memory resource consumption increases
Solution Approach 1:
The patent segments the anomaly detection task into multiple independent random decision trees forming an ensemble. Each tree processes a subset of features and samples, dividing the computational workload and reducing memory requirements while maintaining detection accuracy through the collective decision of multiple trees.
Solution Approach 2:
The patent changes the parameter representation by using random decision trees instead of traditional neural networks, and further optimizes by quantizing continuous parameters into discrete bins. This parameter transformation reduces memory footprint and compute requirements while preserving the essential anomaly detection capability.
2Reliability
If neural networks are used for anomaly detection, then detection capability is improved, but adaptability to varied anomalies deteriorates due to lack of labeled datasets
Solution Approach 1:
The patent inverts the traditional supervised learning approach by using unsupervised learning with random decision trees. Instead of requiring labeled anomaly data to train the model, the system builds random trees on normal data distribution and detects anomalies as deviations from this distribution, enabling adaptability to varied anomaly types without labeled datasets.
Solution Approach 2:
The random decision tree ensemble is self-adaptive to different anomaly types without requiring external labeled data for retraining. The model serves itself by learning the normal data distribution and automatically identifying deviations, providing versatility across different anomaly scenarios.
3Measurement precision
If full-precision data is used for training, then model accuracy is improved, but memory footprint and training time increase
Solution Approach 1:
The patent applies parameter quantization by converting continuous feature values into discrete bins. This transformation reduces the precision requirements from floating-point to integer representations, significantly decreasing memory footprint and training sample requirements while maintaining sufficient accuracy for anomaly detection tasks.
Data Source
AI summary
Systems, apparatuses, and methods include technology that identifies a first dataset that comprises a plurality of data values, and partitions the first dataset into a plurality of bins to generate a second dataset, where the second dataset is a compressed version of the first dataset. The technology randomly subsamples data associated with the first dataset to obtain groups of randomly subsampled data, and generates a plurality of decision tree models during an unsupervised learning process based on the groups of randomly subsampled data and the second dataset.