Data Science and Management Research Laboratory

​

• Evaluation of Demographic Bias in Facial Recognition

  • Led by PI Mustafa Atay, September 2020 - present​

  • Current machine learning algorithms allow for a relatively reliable detection, recognition, and categorization of face images comprised of various age, race, and gender. Algorithms with biased data are bound to produce skewed results. It leads to a significant decrease in the performance of state-of-the-art models when applied to images of gender or race groups. In this research, we study the demographic bias in face and facial expression recognition. We aim to report the machine and deep learning algorithms which are inclined towards demographic bias and the ones which mitigate it besides reporting biased facial image datasets. The current results of this research appeared in Computers journal in 2021, Future Internet journal in 2021 and IEEE SSCI Symposium in 2021 and 2023.

• Evaluation of Machine Learning Models for User Authentication Using Face Biometrics

  • Led by PI Mustafa Atay, October 2019 - present​

  • In this research project, we focused on augmenting traditional username-password standard with face biometrics to enhance user authentication. Using face biometric for user authentication needs an extensive evaluation study to show how reliable and effective it will be under different settings. We elaborated the impact of mask wearing in face recognition where mask wearing is mandated by governments during Covid-19 pandemic. This research study strives to examine and test various configurations along with conventional Machine Learning algorithms which include KNN, SVM, LDA, DT, LR and NB besides deep learning algorithms such as AlexNet, VGG16 and ResNet50 to determine their image classification accuracy. This research reports the most favorable and promising face recognition settings as a potential way to augment the classical username-password standard by increasing their security with facial biometrics. The results of this research appeared in IEEE ICMLA 2020 and 2023, Future Internet 2021 and IEEE SoutheastCon 2023.

• Evaluating Continuous Authentication in Smartphones

  • Led by Co-PI Debzani Deb, October 2019 - present​

  • This research developed a touch-stroke authentication model based on Auxiliary Classifier Generative Adversarial Network (AC-GAN). Given a small subset of a legitimate user’s touch-stroke data during training, the presented AC-GAN model learns to generate a vast amount of synthetic touch-strokes that closely approximate the real touch-strokes, simulating imposter behavior, and then uses both generated and real touch-strokes in discriminating real user from the imposters. Our authentication relies on an architecture where the computationally demanding AC-GAN training takes place on the server-side, and the lightweight mobile side performs the authentication. We trained the presented network on Touch-analytics dataset and evaluated the discriminability with popular performance metrics and loss functions. Our evaluation results suggest that it is possible to achieve comparable authentication accuracies (EER ranging from 2% to 11%) even when the generative model is challenged with a vast number of synthetic data that effectively simulates imposter behavior. Use of AC-GAN also diversifies generated samples and stabilizes training. The findings of this research are significant in the field of biometrics, especially when addressing imposter attacks. The current results of this research appeared in IEEE ICMLA 2020 and ADMI 2021.

• Computational Framework and Data Science for Identification

  • Entire Team, October 2019 - present​

  • Identity is an emerging critical research field due to digitization of life and the accompanying threats. Data science helps by opening new opportunities for identifying and verifying persons in both cyberspace and the physical world. The project involves extending the single-sign-on WebID protocol to allow virtually any authentication technique to be incorporated into the protocol. It proposes authentication using periocular biometrics and active authentication using behavioral biometrics, and mitigating presentation attacks.