Edge Device Sample Evaluation for Model Update Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In edge computing systems, the degradation of machine learning models occurs when noisy labels are used for updating the models, leading to inaccurate classifications and system inefficiencies.
Innovation Solution
A method is implemented at the edge device to evaluate incoming samples by comparing their distributions to pre-defined class distributions, identifying trusted samples with accurate labels and avoiding the use of noisy samples for model updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If noisy labeled samples are used to update the machine learning model, then the model can be continuously updated with new data, but the model performance degrades due to inaccurate labels
Solution Approach 1:
The patent applies preliminary action by evaluating sample quality before model updating. The edge device assesses whether new samples meet quality criteria (distribution consistency, confidence thresholds) before incorporating them into training, preventing noisy labels from degrading model performance while still enabling continuous adaptation with high-quality data
Solution Approach 2:
The patent introduces an intermediary evaluation mechanism between sample acquisition and model updating. This intermediary layer includes quality assessment modules that filter samples based on distribution consistency, confidence scores, and other criteria, acting as a mediator that allows beneficial updates while blocking harmful noisy labels
2Productivity
If all acquired samples are used for model updates, then the productivity of the system increases, but the accuracy decreases due to inclusion of noisy labeled samples
Solution Approach 1:
The patent applies local quality by implementing differentiated processing for different samples. Instead of uniform treatment, the system evaluates each sample's quality characteristics (distribution match, confidence level) and applies appropriate processing - high-quality samples undergo standard updates while low-quality samples are filtered or specially handled, maintaining accuracy while preserving throughput
Solution Approach 2:
The patent uses parameter changes by dynamically adjusting quality thresholds and evaluation criteria based on system state. The edge device can modify confidence thresholds, distribution matching parameters, and selection criteria to optimize the balance between processing throughput and accuracy maintenance under different operating conditions
Data Source
AI summary
Embodiments of the present disclosure provide a method, an electronic device, and a computer program product for evaluating samples. The method includes receiving, at an edge device, a classification model from a cloud server. The method further includes acquiring a sample distribution corresponding to each class in a plurality of classes of the classification model. The method further includes acquiring an input sample which is marked as a first class in the plurality of classes. The method further includes determining whether the input sample conforms to a first sample distribution corresponding to the first class. The method further includes identifying, in response to the input sample conforming to the first sample distribution, the input sample as a trusted sample. The trusted sample indicates that the input sample is correctly marked. By the method, a noise sample with a wrong label can be recognized, thus avoiding model degradation during update.