Sparse Ensembling via Gating Network for Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing unsupervised machine learning ensemble methods require a large number of models, leading to high computing costs due to the need to execute all models for anomaly detection, which is inefficient and resource-intensive.
Innovation Solution
A gating network is trained to select a minimal subset of unsupervised machine learning models based on input data, using binary gate weights to enable or disable models for evaluation, thereby reducing the number of models needed for a final score computation, employing a custom loss function to optimize prediction accuracy and sparsity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large number of unsupervised machine learning models are used in an ensemble for anomaly detection, then the detection accuracy and reliability are improved, but the computing cost and resource consumption increase significantly
Solution Approach 1:
The patent segments the ensemble of unsupervised models into two functional groups: (1) a gating network that processes input data and outputs binary decisions, and (2) a subset of anomaly detection models that are selectively activated based on gating decisions. This segmentation allows the system to maintain multiple models for accuracy while only executing necessary subsets, reducing computing cost.
Solution Approach 2:
The gating network performs preliminary processing of input data before the anomaly detection models are activated. By pre-evaluating the input through the gating network and generating binary gate decisions in advance, the system determines which models need to be executed, avoiding unnecessary model executions and reducing overall computing cost.
2Reliability
If all unsupervised machine learning models are executed for every input, then comprehensive anomaly detection is achieved, but the processing time and computational load increase
Solution Approach 1:
The patent divides the model execution process into two stages: (1) the gating network processes all inputs to generate binary gate decisions, and (2) only the necessary subset of anomaly detection models is executed based on these decisions. This segmentation ensures comprehensive detection where needed while avoiding unnecessary processing, thus reducing processing time.
Solution Approach 2:
Instead of executing all models for every input (excessive action), the system executes only the necessary subset of models determined by the gating network (partial action). The binary gate decisions enable the system to apply partial action by activating only those models that are deemed necessary for the given input, maintaining detection comprehensiveness while reducing processing time.
3Reliability
If multiple anomaly detection models are ensembled using traditional methods, then the performance is improved, but the system complexity and resource requirements increase
Solution Approach 1:
The patent segments the ensemble system into a gating network component and a model subset component. The gating network handles the complexity of model selection by processing inputs and generating binary decisions, while the anomaly detection models focus solely on their detection function. This segmentation reduces system complexity by separating control logic from detection logic.
Solution Approach 2:
The gating network acts as an intermediary between the input data and the anomaly detection models. It mediates the selection process by generating binary gate decisions that determine which models should be activated. This intermediary structure simplifies the overall system by introducing a centralized control mechanism that manages model execution based on input characteristics.
Data Source
AI summary
Techniques are provided for sparse ensembling of unsupervised machine learning models. In an embodiment, the proposed architecture is composed of multiple unsupervised machine learning models that each produce a score as output and a gating network that analyzes the inputs and outputs of the unsupervised machine learning models to select an optimal ensemble of unsupervised machine learning models. The gating network is trained to choose a minimal number of the multiple unsupervised machine learning models whose scores are combined to create a final score that matches or closely resembles a final score that is computed using all the scores of the multiple unsupervised machine learning models.