BACHELOR Assignment
Learning to recognize facial expressions from face images with Convolutional Neural Networks
Type: Bachelor CS
Topic: Machine learning and computer vision
Period: TBD
Student: (Unassigned)
If you are interested please contact :
Introduction:
Automatic recognition of emotions from face images and videos is an important problem in the robotics and human-machine interaction research communities. The reactions/behaviour of robots may change according to the emotions of the human users. The performance of Convolutional Networks are sensitive to the way they are trained and the configuration of hyperparameters used for the model optimization. Stability of performance is needed to deploy effective systems.
Assignment:
This work aims to examine the effect of hyperparameters for the training of Convolutional Neural Networks for emotion recognition from images and videos of faces. The assignment is to train several CNN architectures to predict emotion from face images, evaluating the effect of different hyperparameters on the results of training in terms of validation accuracy, generalization capabilities, stability of results among subsequent runs. The parameters to be evaluated include the learning rate schedule, the batch size, the use of weight decay, the optimizer algorithm, the use and the weight of momentum terms.
References:
[1] J. Hu, L. Shen, S. Albanie, G. Sun and E. Wu, "Squeeze-and-Excitation Networks," IEEE/CVF Conference on CVPR, 2018
[2] S. Li, W. Deng and J. Du, "Reliable Crowdsourcing and Deep Locality-Preserving Learning for Expression Recognition in the Wild," CVPR, 2017
[3] S. L. Smith, P.-J. Kindermans, C. Ying and Q. V. Le, "Don't Decay the learning rate, increase the batch size," ICLR, 2018.
[4] E. Hoffer, I. Hubara and D. Soudry, "Train longer, generalize better: closing thegeneralization gap in large batch training of neural networks," NIPS, 2017.
[5] E. Barsoum, C. Zhang, C. C. Ferrer and Z. Zhang, "Training Deep Networks for Facial Expression Recognition with Crowd-Sourced Label Distribution," ICMI '16 Proceedings of the 18th ACM International Conference on Multimodal Interaction, 2016.