Encoder Neural Network Biometric Authentication Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional authentication systems face challenges in accurately identifying authorized users, leading to false positives and false negatives due to limitations in training encoder neural networks for biometric data processing.
Innovation Solution
A method for training an encoder neural network to generate inter-class and intra-class embedded representations of biometric data samples, which are used to improve authentication systems by adjusting neural network parameters based on loss functions that enhance discrimination and reconstruction fidelity, allowing for more effective user authentication with fewer computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional encoder neural networks are used for biometric authentication, then the system can process biometric data, but it produces false positives and false negatives due to insufficient discrimination between inter-class and intra-class variations
Solution Approach 1:
The embedded representation is segmented into two distinct components: inter-class embedded representation (capturing identity-specific features) and intra-class embedded representation (capturing identity-agnostic variations). This segmentation allows the system to separately optimize for discrimination between different identities and robustness to variations within the same identity, thereby improving both reliability and measurement precision simultaneously
Solution Approach 2:
Different parts of the embedded representation are assigned different functional qualities: the inter-class component is optimized for discrimination precision (identifying who the person is), while the intra-class component is optimized for robustness to environmental variations (handling different lighting, angles, expressions). This local quality differentiation resolves the contradiction by allowing each component to excel at its specific function
2Reliability
If more training data and computational resources are used to improve authentication accuracy, then false positives and false negatives decrease, but the training time and computational cost increase significantly
Solution Approach 1:
By segmenting the loss function into inter-class loss and intra-class loss components, the training process can more efficiently converge. The inter-class loss focuses on separating different identities, while the intra-class loss focuses on compacting variations within the same identity. This segmented approach achieves better authentication accuracy with fewer training iterations compared to conventional unified loss functions
Solution Approach 2:
The patent introduces a specific parameterization of the embedded representation that separates identity-specific and identity-agnostic features. This parameter change enables the model to learn more effective representations faster, reducing the number of training iterations needed to achieve high authentication accuracy
3Measurement precision
If the encoder neural network is trained to improve discrimination between identities, then authentication accuracy improves, but the system complexity and training difficulty increase
Solution Approach 1:
The complexity of achieving high discrimination precision is segmented into two manageable tasks: learning inter-class discriminative features and learning intra-class invariant features. This segmentation makes the training process more tractable and the model architecture more interpretable, reducing the perceived complexity while maintaining high discrimination precision
Solution Approach 2:
The encoder neural network is designed to simultaneously serve multiple functions: it extracts identity-specific features for discrimination, captures identity-agnostic features for robustness, and generates embedded representations that can be used for both authentication and enrollment. This multi-functionality reduces overall system complexity by consolidating multiple operations into a single unified model
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an encoder neural network having multiple encoder neural network parameters. The encoder neural network is configured to process a biometric data sample in accordance with current values of encoder neural network parameters to generate as output an embedded representation of the biometric data sample. The embedded representation includes: (i) an inter-class embedded representation, and (ii) an intra-class embedded representation that is different than the inter-class embedded representation.