IoT Device Feature Selection for Federated Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training machine learning models on IoT devices is challenging due to resource constraints and heterogeneity among devices, which affects collaborative learning and delays the training process, especially when privacy concerns require resource status to be kept private.
Innovation Solution
A method where client computing devices determine suitable sensors for data collection based on measurement specifications and resource usage, allowing them to dynamically adjust settings and select features for training without revealing resource information, enabling efficient training of machine learning models in heterogeneous environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If federated learning is used to train machine learning models on IoT devices, then collaborative learning capability is improved, but resource constraints and heterogeneity cause training delays and privacy concerns
Solution Approach 1:
The patent segments the training process by dividing IoT devices into groups based on their resource capabilities and sensor configurations. Each group trains on subsets of features appropriate to their capabilities, allowing simultaneous training across multiple groups without mutual interference, thus reducing overall training delay while maintaining collaborative learning benefits
Solution Approach 2:
The patent implements dynamic feature selection where devices adaptively choose which features to collect and process based on their current resource availability and sensor status. This dynamic adjustment allows devices to optimize their participation in federated learning according to real-time conditions, reducing training delays caused by resource constraints
2Measurement precision
If devices collect and process more features for training, then model accuracy is improved, but resource consumption increases
Solution Approach 1:
The patent applies local quality by allowing different IoT devices to collect and process different subsets of features based on their specific sensor configurations and resource capabilities. Each device optimizes its local feature set to achieve adequate model accuracy while minimizing its own resource consumption, rather than all devices uniformly collecting all possible features
Solution Approach 2:
The patent changes the parameter of feature selection dynamically based on device resource status. Devices can adjust which features are collected, the sampling rate, and the complexity of local processing according to available computational resources, energy levels, and memory constraints, thereby maintaining acceptable model accuracy while adapting resource consumption to available capacity
3Productivity
If client resource status is shared for optimization, then training efficiency is improved, but privacy is compromised
Solution Approach 1:
The patent introduces an intermediary mechanism where devices share aggregated or anonymized capability information rather than detailed resource status. The coordinating device uses this intermediate information to assign features and group devices without exposing sensitive private information about individual device resource constraints, maintaining both training efficiency and privacy
Data Source
AI summary
A method by a client computing device having one or more sensors for collecting data, includes obtaining information identifying a first set of measurable features, each feature being associated with a measurement specification. For each feature of the first set of measurable features, determining whether there is at least one sensor of the one or more sensors satisfying the associated measurement specification. If there is at least one sensor of the one or more sensors satisfying the associated measurement specification, estimating a resource usage. Determining a first subset of the first set of measurable features and sending information identifying the first subset of the first set of measurable features. When the client computing device is determined to belong to the first group of computing devices, performing training of the machine learning model using the first group of computing devices.