Decentralized Autoencoder for Minority Class Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing approaches for detecting sleeping cells in communication networks face challenges such as imbalanced datasets, privacy concerns, high data volume collection, latency issues, and large training datasets, which affect the performance and efficiency of network utilization.
Innovation Solution
A decentralized autoencoder method is employed to manage imbalanced datasets by selecting and balancing data from communication devices, using RAN data to determine class distributions, and averaging model parameters to reduce training time and network footprint, while preserving privacy and reducing data transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a decentralized autoencoder is used to detect minority classes from imbalanced datasets, then detection accuracy of minority classes is improved, but the complexity of managing distributed datasets and model training increases
Solution Approach 1:
The system segments the training process by having each communication device train on its own local imbalanced dataset independently, then aggregates only the learned parameters rather than raw data. This segmentation allows accurate minority class detection while reducing the complexity of centralized data management.
Solution Approach 2:
The master communication device acts as an intermediary that coordinates the decentralized training process. It collects information about local dataset compositions from participating devices, manages the aggregation of learned parameters, and orchestrates the overall training workflow without requiring direct peer-to-peer communication between all devices.
2Productivity
If communication devices report detailed data composition statistics to a master device, then the master can select appropriate devices for training, but the signaling overhead and network usage increase
Solution Approach 1:
The system extracts only the essential information needed for device selection - the composition statistics of local datasets (number of samples per class) - and transmits this extracted information to the master device. This selective extraction reduces signaling overhead while still enabling effective device selection based on dataset characteristics.
3Measurement precision
If a large training dataset is used to ensure sufficient minority class samples, then detection accuracy improves, but training time and network footprint increase
Solution Approach 1:
The system allows each communication device to contribute its local expertise by training on its own local dataset with its own specific minority class samples. This local quality approach ensures that the diverse characteristics of minority classes across different devices are captured, achieving accurate detection without requiring a single large centralized dataset.
Solution Approach 2:
The system merges the learned parameters from multiple decentralized training processes into a unified model. By combining the knowledge gained from different local datasets through parameter aggregation, the system achieves comprehensive minority class detection capability equivalent to training on a large unified dataset, but with reduced training time and network footprint.
Data Source
AI summary
A method performed by a first communication device for managing a decentralized autoencoder for detection or prediction of a minority class from an imbalanced dataset is provided. The method includes signalling a message to communication devices including a set of parameters for the decentralized autoencoder; and receiving a message from a communication device providing information including an amount of local samples and a distribution of labels in the local samples of instances of local majority class samples and/or the local minority class samples. The method further includes computing a computed number of samples and a computed distribution of labels for aggregated local majority class samples and aggregated local minority class samples; and selecting a set of communication devices to include in the decentralized autoencoder based on the communication devices that can satisfy the computed number of samples and the computed distribution of labels.