Autoencoder Drift Detection for Image Classification Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning models deployed in production environments are negatively impacted by data drift and outliers, leading to inaccurate predictions due to differences between training and practical data, sensor inefficiencies, and transient changes.
Innovation Solution
Implement an autoencoder on edge devices to reconstruct images, capture reconstruction errors, and use affinity propagation for clustering to detect outliers and data drift, associating this information with model classifications to refine the model on a cloud server.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained with labelled data in cloud environments, then model training performance is improved, but the models become vulnerable to data drift and outliers in production environments
Solution Approach 1:
The patent applies preliminary action by detecting data drift and outliers before they significantly impact model performance. The system continuously monitors input data distributions and identifies anomalies in advance, allowing for proactive model updates that maintain prediction accuracy despite changing production conditions.
Solution Approach 2:
The patent implements feedback mechanisms where production data is continuously fed back into the system to update model parameters. This closed-loop approach allows the model to adapt to data drift and outliers by incorporating real-world performance information, thereby maintaining reliability while improving adaptability.
2Speed
If edge devices perform analytics locally, then data processing speed is improved, but computing capability is limited
Solution Approach 1:
The patent applies segmentation by dividing analytics tasks between edge devices and cloud servers. Edge devices perform lightweight local analytics and data collection, while complex model training and heavy computational analytics are segmented and executed in the cloud environment, optimizing the balance between processing speed and capability.
Solution Approach 2:
The patent uses cloud servers as intermediaries that receive data from edge devices, perform complex analytics and model training, then return processed results. This intermediary approach allows edge devices to maintain fast local processing for simple tasks while leveraging cloud power for complex computations.
3Productivity
If more analytics are performed on edge systems, then local processing capability is improved, but device resource consumption increases
Solution Approach 1:
The patent applies partial action by having edge devices perform only the necessary minimal analytics locally (such as data collection and simple filtering) while deferring complex computational tasks to the cloud. This partial execution approach maintains adequate local productivity for immediate needs while significantly reducing energy consumption compared to performing all analytics locally.
Data Source
AI summary
A system and a method for updating a Machine Learning (ML) model are described The method involves capturing reconstruction errors associated with reconstruction of images by a pre-trained autoencoder. Data points representing the reconstruction errors are clustered using affinity propagation. A preference value used by the affinity propagation for determining similarity between the data points is dynamically set through linear regression. Outliers and data drifts are determined from clusters of the data points. Classification output of the ML model is associated with the outliers and the data drift, for refinement of the ML model over a device hosting a training environment.