October 22, 2021

About the speaker

Dr. Dimah Dera (Member IEEE) is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Texas Rio Grande Valley. She received her Ph.D. in Electrical and Computer Engineering, M.A. in Mathematics and M.S. in Electrical and Computer Engineering from Rowan University, USA and B.S. in Biomedical Engineering from Damascus University, Syria. She was a postdoctoral fellow with the Department of Electrical and Computer Engineering at Rowan University from 2020 to 2021. She won the Best Student Paper Award at the 2019 IEEE International Workshop on Machine Learning for Signal Processing (MLSP’19) and the Runner-up Best Paper Award at the 2015 IEEE International Conference on Bioinformatics and Biomedicine (BIBM’15). She is the recipient of the NJ Health Foundation Research Grant Award (2021), the IEEE Philadelphia Sections Benjamin Franklin Key Award (2021), the STEM Innovator to Watch award from the NJ Tech Council (2018), the Graduate Research Excellence Award from Rowan University (2017), the National Science Foundation iREDEFINE Professional Development Award (2017) and the ACM SIGHPC/Intel Computational and Data Science Fellowship (2016). Her research interests are in Big Data Analytics, Machine Learning, with a special focus on Bayesian deep learning, adversarial learning, statistical tracking and optimization.

Towards Machine Self-Awareness – A Bayesian Framework for Uncertainty Propagation in Deep Neural Networks

Deep neural networks (DNNs) have surpassed human-level accuracy in various fields, including object identification, face recognition, voice recognition and even painting. Deep learning techniques hold the promise of emerging technologies, such as self-driving cars and autonomous unmanned aircraft systems, smart cities infrastructure, personalized treatment in medicine, and cybersecurity. However, unlike Humans who have a natural cognitive intuition for probabilities, DNN systems, being inherently deterministic, are unable to evaluate their confidence in the decisions. To truly deserve its name, an artificial intelligence system must be aware of its limitations and have the capacity for insightful introspection. This seminar will introduce new deep learning methods that are able to quantify their uncertainty in the decision and self- assess their performance, are robust to adversarial attacks and can even expose an attack from ambient noise. The main contribution of this work is establishing the theoretical and algorithmic foundations of uncertainty or belief propagation through complex deep learning models by adopting powerful frameworks from density tracking in non-linear and non- Gaussian dynamical systems.

Bayesian inference provides a principled approach to reason about model confidence or uncertainty by estimating the posterior distribution of the unknown parameters given the observed data. The challenge in DNNs is the multi-layer stages of non-linearities in deep learning models, which makes propagation of high-dimensional distributions mathematically intractable. Drawing upon powerful statistical frameworks for density propagation in non-linear and non-Gaussian dynamical systems, we introduce Tensor Normal distributions as priors over the network parameters and derive a first-order Taylor series mean-covariance propagation framework. We subsequently extend this first-order approximation to an unscented framework that propagates sigma points through the model layers. The unscented framework is shown to be accurate to at least the second-order approximation of the posterior distribution. The uncertainty in the output decision is given by the propagated covariance of the predictive distribution. Furthermore, we show that the proposed framework performs an automatic logit squeezing, which leads to significantly enhanced robustness against noise and adversarial attacks. Experimental results on benchmark datasets, including MNIST, CIFAR-10, real-world synthetic aperture radar (SAR) and Brain tumor segmentation (BraTS 2015), demonstrate: 1) superior robustness against Gaussian noise and adversarial attacks; 2) self-assessment through predictive confidence that decreases quickly with increasing levels of ambient noise or attack; and 3) an ability to detect a targeted attack from ambient noise.