Graph-Regularized Multi-Sensor Fusion for Missing and Noisy Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-sensor fusion methods are unable to effectively handle missing and noisy labels, which are prevalent in practical application scenarios, especially in data-limited scenarios, and fail to provide a unified framework for semi-supervised learning and multi-sensor fusion.
Innovation Solution
The Graph regularized AutoFuse (GAF) method employs a joint optimization formulation for learning sensor-specific autoencoders and classification, utilizing a graph-based approach to propagate and correct labels, combining latent representations from multiple sensors and integrating known label information to improve classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling of huge training data set is performed, then classification accuracy can be improved, but time consumption and cost increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training autoencoder models on unlabeled data to learn effective feature representations before performing classification. This allows the system to leverage abundant unlabeled data to prepare robust models that require fewer labeled examples for fine-tuning, thereby reducing the time and cost of manual labeling while maintaining classification accuracy.
Solution Approach 2:
The patent implements self-service through self-supervised learning mechanisms where the model automatically generates pseudo-labels for unlabeled data and performs self-training. The autoencoder framework enables the system to learn from its own reconstructed data, eliminating the need for extensive manual labeling while achieving accurate classification through iterative self-improvement.
2Loss of time
If semi-supervised learning methods are used, then labeling cost is reduced, but ability to handle noisy and missing labels is insufficient
Solution Approach 1:
The patent introduces an intermediary mechanism through the autoencoder framework that acts as a bridge between labeled and unlabeled data. The autoencoder learns robust latent representations from unlabeled data that can tolerate noise and missing labels, then uses these representations to guide the classification process. This intermediary representation layer enables the system to handle noisy and missing labels effectively while maintaining low labeling costs.
Solution Approach 2:
The patent applies parameter changes by modifying the loss function to include regularization terms that specifically address noisy and missing labels. The framework dynamically adjusts weighting parameters during training to emphasize reliable labels and downweight potentially noisy ones, enabling effective semi-supervised learning even when labels are imperfect while keeping labeling costs low.
3Measurement precision
If existing multi-sensor fusion methods are applied, then classification performance can be achieved, but they fail to handle missing and noisy labels effectively
Solution Approach 1:
The patent applies segmentation by dividing the multi-sensor fusion process into separate autoencoder modules for each sensor type. Each sensor-specific autoencoder independently processes its data and handles its own missing or noisy labels, then the learned representations are fused at the latent space level. This segmented approach allows each module to specialize in handling its sensor's specific quality issues while maintaining overall classification performance.
Solution Approach 2:
The patent implements composite materials by creating a composite learning framework that combines multiple autoencoder models (each trained on different sensor data) into a unified multi-sensor fusion system. This composite structure integrates diverse sensor representations with different noise and missingness patterns, creating a robust ensemble that leverages the strengths of each sensor while compensating for their individual weaknesses through the fused latent representation.
Data Source
AI summary
This disclosure relates to a method and system for multi-sensor fusion in the presence of missing and noisy labels. Prior methods for multi-sensor fusion do not estimate and correct labels for learning effective models in semi-supervised learning methods. Embodiments of the present disclosure provides a method for learning robust sensor-specific autoencoder based fusion model by utilizing a graph structure to perform label propagation and correction. In the disclosed Graph regularized AutoFuse (GAF) method latent representation for each sensor is learnt using the sensor-specific autoencoders. Further these latent representations are combined and fed to a classifier for multi-class classification. The disclosure presents a joint optimization formulation for multi-sensor fusion where label propagation and correction, sensor-specific learning and classification are executed together.