Edge Neural Network Activation Data Filtering for Bandwidth Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for updating machine learning models with new data from edge devices are inefficient, as they either overload edge devices with processing or waste bandwidth by transmitting all data to servers for evaluation.
Innovation Solution
An edge device runs a reduced neural network to determine activation data for new data points and sequentially transmits this data to a server, which compares it to previously encountered data, instructing the edge to stop sending when the data is within an expected range, thus only transmitting relevant data for further training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the edge device determines whether data points are suitable for training, then data transmission efficiency is improved, but the computational burden on the edge device increases
Solution Approach 1:
The system segments the data filtering task into two parts: the edge device performs initial processing by running data through a neural network and transmitting activation data, while the server performs the final suitability determination by comparing activation data against training criteria. This segmentation allows the edge device to offload computationally intensive tasks while still contributing to efficient data transmission.
Solution Approach 2:
The edge device performs a partial evaluation by computing activation data through the neural network, which is less computationally demanding than full training model determination. This partial action at the edge device, combined with server-side verification, achieves efficient data filtering without overwhelming the edge device's computational resources.
2Measurement precision
If all data points are transmitted from the edge device to the server, then data suitability determination accuracy is improved, but data transmission bandwidth is wasted
Solution Approach 1:
The system extracts only the essential information needed for suitability determination by transmitting activation data from the neural network's intermediate layers rather than transmitting all raw input data. This extraction approach maintains the server's ability to accurately assess data suitability while significantly reducing the volume of data that needs to be transmitted over the network.
Solution Approach 2:
Instead of transmitting the original raw data, the system transmits a transformed representation (activation data) that captures the essential features needed for suitability determination. This copying approach allows the server to perform accurate evaluation using a more compact data representation, reducing bandwidth consumption while maintaining determination accuracy.
Data Source
AI summary
Systems and methods relating to machine learning. An edge device runs a new data point on a first neural network and determines activations on the layers within that neural network. The first neural network is a fully trained network based on a second neural network on a server. The activation data for the various layers in the first neural network are, starting with the output layer, sequentially transmitted to the server. The server continuously receives this activation data and continuously compares it with previously encountered activation data for the second neural network. If the received activation data is within an expected range, then the edge device is instructed to stop sending activation data. Otherwise, the server continues to receive the activation data for the other layers until the new data point is received by the server or the activation data is within the expected range of previously encountered activation data.