Tiny Face Recognition Models Using Genetic Architecture Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for creating face recognition systems in resource-limited settings, such as microcontroller-based edge devices, are inefficient and lack accuracy due to the complexity of existing methods like Neural Architecture Search (NAS) and the challenge of handcrafting tiny models for diverse datasets.
Innovation Solution
A method and system using a genetic algorithm to create a tiny machine learning model for face recognition, which involves preprocessing images, creating a master dataset, applying dimensionality reduction methods, and iteratively training gene structures to optimize model accuracy and size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large, full-fledged DNN models are used for face recognition, then model accuracy is improved, but device resource requirements increase
Solution Approach 1:
The patent extracts only the essential components needed for face recognition by removing unnecessary layers and operations from full-fledged DNN models. The genetic algorithm systematically prunes the network architecture to retain only critical feature extraction capabilities, creating a compact model that fits resource-limited devices while preserving recognition accuracy.
Solution Approach 2:
The patent changes key parameters of the neural network including reducing the number of layers, decreasing filter counts, optimizing learning rates, and adjusting batch sizes. These parameter modifications are performed iteratively through the genetic algorithm to achieve optimal balance between model size and recognition accuracy for embedded systems.
2Measurement precision
If existing NAS methods are used to create customized models, then model accuracy is improved, but computing resource requirements and search time increase
Solution Approach 1:
The patent segments the model creation process into distinct phases: preprocessing, dimensionality reduction, genetic algorithm-based architecture search, and iterative training. This segmentation allows each phase to be optimized independently, reducing overall computational complexity while maintaining accuracy through focused optimization at each stage.
Solution Approach 2:
The patent performs preliminary actions by pre-processing images and creating dimensionality-reduced datasets before the main model training. This preliminary dimensionality reduction using techniques like PCA or LDA decreases the input data complexity, making subsequent model training more efficient and reducing the computational burden during the genetic algorithm search.
3Quantity of substance
If tiny models with limited layers are used for resource-limited devices, then device resource requirements are reduced, but model accuracy deteriorates
Solution Approach 1:
The patent systematically adjusts parameters such as the number of convolutional layers, filter depths, pooling operations, and activation functions to find the optimal configuration for tiny models. The genetic algorithm evaluates multiple parameter combinations to identify the configuration that achieves maximum accuracy within the constraints of limited model size.
Solution Approach 2:
The patent implements feedback mechanisms where the model performance is continuously evaluated during the genetic algorithm iterations. Accuracy metrics from validation sets provide feedback that guides the selection and evolution of model architectures, allowing the system to learn from performance results and progressively improve tiny model accuracy through iterative refinement.
Data Source
AI summary
Nowadays, intelligent analysis of image data becomes increasingly pivotal in various domains such as robotics, autonomous vehicles, security, and first level disease detection. Existing techniques available for image processing requires large computing structure and suffers from high search time for searching networks connected in complex manner. Present disclosure provides method and system for creating face recognition system in tiny machine leaning (ML) model using genetic algorithm. The system first receives training images which are then pre-processed to obtain facial images. Then, system creates master dataset and reduced datasets. Further, system creates plurality of gene structures using tiny neural network. Each gene structures is then trained with created reduced dataset. Furthermore, performance of each trained gene structure is evaluated based on accuracy score and model size which helps in determining best gene structure among the plurality of gene structures. Finally, tiny new ML model is created based on best gene structure.