Publication Date

Fall 2020

Degree Type


Degree Name

Master of Science (MS)


Computer Engineering


Harry Li


adaptive learning, Computer Vision, FaceNet, Facial Recognition, Machine Learning, ResNet

Subject Areas

Computer engineering; Computer science


This research describes the adaptive learning technique for facial recognition. It is a common practice in convolutional neural network(CNN) based facial recognition to save its trained result on a large dataset and then load and apply it to ongoing facial recognition tasks. This generally used method lacks adaptation, and the ongoing evolution of new knowledge poses a key technical challenge. In this research, we propose a continued learning technique to incorporate new knowledge derived in each facial recognition process. A positive recognition with confidence score is assigned, and the image associated with this confidence is added to the image dataset for ongoing training. Pre-trained CNN on a similar small dataset serves as the starting point for this ongoing training technique, which leads to a significant reduction in the training time and enhancement of the recognition rate. This research is inspired by the evolutionary adaptive learning talk given by Dr. Harry Li in the 2019 SiliconValley AI Event. This research is conducted to provide proof-of-concept groundwork to demonstrate the feasibility of continued learning and adaptation while executing the FaceNet/ResNet facial recognition algorithm. In this research, the proof-of-the-concept algorithm is demonstrated on a simple feed-forward neural network, then tested with an adaptive face technique to demonstrate the learning acceleration from the adaptive process. Experiments confirm the adaptive learning on FaceNet and ResNet as per the proof-of-the-concept by reducing the number of epochs required to reach a convergence point by approximately 50%. This states the use of adaptive learning techniques in software that require identifying images of aging people concerning time.