Neural Network Models for API Gateway Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In complex gateway architectures, anomaly detection becomes unmanageable due to the overwhelming number of service requests, making it difficult for engineering teams to identify and resolve issues in a timely manner.
Innovation Solution
The implementation of a system utilizing multiple small neural networks that can be trained and retrained quickly, processing streaming data in parallel to detect anomalies in real-time, with each network associated with a specific service and using metadata to determine the features for training and anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a single deep neural network is used for anomaly detection, then the system can process large amounts of data, but the training time and processing power requirements become unmanageably high
Solution Approach 1:
The patent divides a single deep neural network into multiple smaller neural networks, each responsible for detecting anomalies in specific services. This segmentation reduces the computational burden on each individual network, enabling faster training and inference while maintaining the ability to process large volumes of data across all services collectively.
Solution Approach 2:
The patent introduces a new dimension of organization by grouping neural networks according to services rather than using a single monolithic network. This dimensional change allows parallel processing of data from different services, effectively increasing data processing capacity while reducing the time required to train each individual network model.
2Quantity of substance
If a single deep neural network is used for anomaly detection, then the system can process large amounts of data, but the processing power requirements become excessive
Solution Approach 1:
The patent segments the processing workload across multiple smaller neural networks, each handling a subset of services. This distribution reduces the processing power required by any single network while maintaining the overall data processing capacity of the system through parallel operation of multiple networks.
Solution Approach 2:
The patent merges multiple specialized neural networks into a unified anomaly detection system that collectively processes data from all services. This merging allows the system to maintain high data processing capacity while each individual network operates within reasonable processing power constraints.
3Reliability
If traditional anomaly detection methods are used in complex gateway architectures, then engineering teams can monitor error messages, but the effort becomes unmanageable with large numbers of services
Solution Approach 1:
The patent implements self-service anomaly detection by training neural networks to automatically identify and flag anomalous patterns in service data. This eliminates the need for manual monitoring of error messages by engineering teams, allowing the system to reliably detect anomalies across large numbers of services without increasing operational complexity.
Solution Approach 2:
The patent replaces manual anomaly detection mechanisms with automated neural network-based detection. This substitution transforms the mechanical process of human engineers reviewing error messages into an automated system that can handle complex gateway architectures with numerous services, maintaining reliability while reducing operational burden.
4Quantity of substance
If a deep neural network with backpropagation is used, then the network can learn from large quantities of data, but retraining with additional data remains time-consuming
Solution Approach 1:
The patent segments the training process across multiple smaller neural networks, each trained on data from specific services. This allows for more efficient retraining when new data becomes available, as individual networks can be updated independently and in parallel, reducing the overall retraining time while still processing large volumes of training data.
Data Source
AI summary
API gateway log data of HTTP response information is collected and aggregated for each API service. The log data is streamed and formed into a feature vector. A model training module trains a neural network model based on the feature vector and model metadata. The model training module also retrains the neural network model based on a new feature vector. Neural network model artifacts are stored and used to detect anomalous patterns and generate alerts of the detection.