Learning Data Processor for IoT Network Feature Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In Low Power and Lossy Networks (LLNs), such as IoT networks, the complexity and constraints on resources like energy, memory, and bandwidth hinder the effective use of Learning Machines (LMs) due to inefficient classic approaches and the inability to process vast amounts of data for predicting network behavior, leading to increased computational complexity and unnecessary traffic overhead.
Innovation Solution
A Learning Data Processor (LDP) is introduced to collect and process data from various sources, compute relevant features, and dynamically adjust transmission rates, while performing data completion and predictive analysis to provide necessary information to LMs, thereby reducing traffic and computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If Learning Machines are deployed in LLNs to predict network behavior, then network management capability is improved, but computational complexity and energy consumption increase beyond the capacity of individual LLN nodes
Solution Approach 1:
The patent segments the learning machine functionality into two parts: a training phase that occurs externally (offline) and a deployment phase that occurs in the LLN (online). The heavy computational workload of model training is separated from the resource-constrained LLN nodes, while only the lightweight inference phase is deployed to the nodes. This segmentation allows advanced network management capabilities without overburdening individual LLN devices.
Solution Approach 2:
The patent introduces an intermediary training system that acts as a mediator between the available network data and the learning machine model. This intermediary performs the computationally intensive training offline using aggregated data from multiple LLN nodes, then distributes the trained model to the nodes. This intermediary approach enables complex learning capabilities while protecting resource-constrained LLN nodes from excessive computational demands.
2Device complexity
If classic approaches are used to process network data in LLNs, then simplicity is maintained, but traffic overhead increases due to inefficient data processing
Solution Approach 1:
The patent applies preliminary action by performing data aggregation and feature extraction before the actual learning inference takes place. The LLN nodes pre-process and aggregate their local data, then send only the essential aggregated information to the learning machine for inference. This preliminary processing reduces the volume of data that needs to be transmitted across the network, thereby reducing traffic overhead while maintaining processing effectiveness.
3Measurement precision
If more data is collected from LLN nodes for learning machine training, then model accuracy is improved, but network bandwidth consumption increases
Solution Approach 1:
The patent extracts only the most relevant features from the raw network data collected by LLN nodes. Instead of transmitting and processing all available data, the system identifies and extracts key features that are most predictive of network behavior. This feature extraction approach maintains high prediction accuracy by focusing on the most informative data elements while significantly reducing the bandwidth required for data transmission and processing.
Data Source
AI summary
In one embodiment, a learning data processor determines a plurality of machine learning features in a computer network to collect. Upon receiving data corresponding to the plurality of features, the learning data processor may aggregate the data, and pushes the aggregated data for select features to interested learning machines associated with the computer network.