Cloud-Device Collaborative Anomaly Detection for Resource Usage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computing devices face performance and energy consumption issues due to applications running abnormally, which can lead to reduced user experience, excessive energy drain, and potential device damage from excessive heating.
Innovation Solution
A cloud-device collaborative method using machine learning-based abnormality detection models to monitor and detect anomalies in resource usage, allowing for timely actions to mitigate these issues and improve user experience and device performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If applications run abnormally on computing devices, then user experience deteriorates and energy consumption increases, but traditional detection methods are insufficient to identify these anomalies in real-time
Solution Approach 1:
The system performs preliminary actions by collecting resource usage data and training abnormality detection models in advance before actual anomalies occur. The models are pre-trained on historical data from multiple devices to recognize abnormal patterns, enabling real-time detection without requiring complex runtime analysis of each anomaly instance.
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring resource usage data from computing devices and using this feedback to update and refine the abnormality detection models. The models learn from actual device performance data, improving their ability to detect anomalies while adapting to different device characteristics and usage patterns.
2Measurement precision
If cloud-device collaborative detection is implemented, then detection accuracy improves, but system complexity increases
Solution Approach 1:
The system segments the anomaly detection functionality into two parts: resource usage data collection on individual computing devices, and model training and updating on remote servers. This segmentation allows each device to operate independently with simple local agents while leveraging collective data from multiple devices for accurate model training, reducing individual device complexity while maintaining high detection accuracy.
Solution Approach 2:
The system introduces an intermediary component - the abnormality detection model - that acts as a mediator between raw resource usage data and anomaly detection decisions. The model processes and interprets complex multi-device data patterns, enabling accurate anomaly detection without requiring direct complex interactions between devices or manual analysis of raw data.
3Difficulty of detecting and measuring
If real-time monitoring of resource usage is performed, then abnormality detection capability improves, but energy consumption increases
Solution Approach 1:
The system applies partial action by implementing lightweight agents on computing devices that only collect essential resource usage data rather than performing full analysis. The heavy computational work of anomaly detection is performed partially on-device using pre-trained models and partially on-remote servers, reducing the energy burden on individual devices while maintaining comprehensive detection capability.
Data Source
AI summary
A computing device receives, from a remote device, one or more abnormality detection models for detecting anomaly of one or a plurality of applications with respect to one or more resource usage, obtains resource usage data associated with each application, determines, using the one or more abnormality detection models, whether each application has anomaly in a resource usage, and takes an action in response. The abnormality detection model is built using a machine learning technique at the remote device based on data collected from multiple devices. The resource usage data includes information about applications used by the computing device, e.g., usage or energy consumption of one or more hardware components or one or more services accessed by an application on the device. An abnormality detection result may include likelihoods of an application falling into an abnormality level in a plurality of abnormality levels.