Convolutional Neural Network Hyperparameters optimization for Facial Emotion Recognition
Adrian Vulpe-Grigorasi, Ovidiu Grigore
Abstract
This paper presents a method of optimizing the hyperparameters of a convolutional neural network in order to increase accuracy in the context of facial emotion recognition. The optimal hyperparameters of the network were determined by generating and training models based on Random Search algorithm applied on a search space defined by discrete values of hyperparameters. The best model resulted was trained and evaluated using FER2013 database, obtaining an accuracy of 72.16%.
Topics & Concepts
HyperparameterComputer scienceConvolutional neural networkArtificial intelligencePattern recognition (psychology)Context (archaeology)Machine learningArtificial neural networkPaleontologyBiologyFace and Expression RecognitionEmotion and Mood RecognitionFace recognition and analysis