Active Learning Neural Network Uncertainty Sampling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Annotating large-scale image datasets for visual object recognition using neural networks is expensive and time-consuming, requiring extensive human effort to label important images.

Innovation Solution

An active learning method that uses an uncertainty measure of features and reconstruction errors to selectively annotate the most informative images, reducing the need for extensive human annotation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If large-scale image datasets are used for training neural networks, then classification accuracy is improved, but annotation cost and time consumption increase significantly

Engineering Contradiction:
Improveclassification accuracyVSAvoidannotation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by selecting only a subset of most informative samples for annotation rather than annotating the entire dataset. The uncertainty sampling strategy identifies samples that provide maximum information gain, allowing the system to achieve high classification accuracy with a fraction of the annotation effort required for complete dataset labeling.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system implements feedback through iterative retraining cycles where the neural network is trained on initially annotated samples, then used to identify new informative samples via uncertainty sampling. This closed-loop feedback mechanism continuously improves classification accuracy by selecting samples that maximize information gain based on the current model state.

Inventive Principle:
Principle #23Feedback

2Measurement precision

If more samples are annotated and added to training set, then classification accuracy is improved, but computational resources and energy consumption increase

Engineering Contradiction:
Improveclassification accuracyVSAvoidCPU usage and power consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent reduces energy consumption by processing only a partial subset of samples through the expensive annotation and retraining pipeline. By using uncertainty sampling to identify only the most informative samples, the system avoids the computational waste of annotating and retraining on samples that would provide minimal accuracy improvement.

Inventive Principle:
Principle #16Partial or excessive action

3Ease of manufacture

If uncertainty sampling based only on top layer outputs is used, then implementation is simple, but classification accuracy is limited

Engineering Contradiction:
Improveimplementation simplicityVSAvoidclassification accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent extends the uncertainty sampling from a single dimension (top layer output margins) to multiple dimensions by incorporating intermediate layer feature uncertainties. This dimensional expansion captures richer information about model confidence and sample informativeness across different levels of feature abstraction, thereby improving classification accuracy while maintaining the simplicity of the sampling framework.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3545470B1Method for training neuron network and active learning system
Publication Date: 2025.02.19 MITSUBISHI ELECTRIC CORP
  • EP3545470B1 patent drawingFigure 1A
  • EP3545470B1 patent drawingFigure 1B
  • EP3545470B1 patent drawingFigure 1C

AI summary

A method for training a neuron network using a processor in communication with a memory includes determining features of a signal using the neuron network, determining an uncertainty measure of the features for classifying the signal, reconstructing the signal from the features using a decoder neuron network to produce a reconstructed signal, comparing the reconstructed signal with the signal to produce a reconstruction error, combining the uncertainty measure with the reconstruction error to produce a rank of the signal for a necessity of a manual labeling, labeling the signal according to the rank to produce the labeled signal; and training the neuron network and the decoder neuron network using the labeled signal.