Arezoo Rajabi

Senior Quantitative Data Analytic Specialist AI/ML

My Publications

Privacy Preserving Reinforcement Learning Beyond Expectation (CDC 2022):

In this paper, we incorporated cumulative prospect theory (CPT) into the objective of a reinforcement learning (RL) problem to quantify risk and used differential privacy to keep decision making hidden from external parties.

Paper

Game of Trojans: A Submodular Byzantine Approach (Under Review) : In this paper, we provide an analytical characterization of adversarial capability and strategic interactions between the adversary and Trojan model detection mechanism.

Paper

Adversarial Images Against Super-Resolution Convolutional Neural Networks for Free (PoPETS 2022) In this work, we hypothesize and empirically show that adversarial examples learned over CNN image classifiers can survive processing by SRCNNs and lead them to generate poor quality images that are hard to classify correctly.

Paper

Resilience Against Data Manipulation in Distributed Synchrophasor-Based Mode Estimation (IEEE Transactions on Smart Grid): In this paper, we introduce a mechanism, inspired by byzantine fault tolerance, for making distributed alternating direction of multipliers method of mode estimation resilient against data manipulation attacks.

Paper

On the (Im)Practicality of Adversarial Perturbation for Image Privacy (PoPETS2021): In this paper, we propose two approaches for gnerating effective transferable perturbations, called – (i) learned universal ensemble perturbations (UEP), and (ii) k-randomized transparent image overlays (k-RTIO) that are semantic adversarial perturbations.

Paper

Toward Metrics for Differentiating Out-of-Distribution (OOD) Sets (ECAI 2020): In this paper, we propose three novel computationally-efficient metrics for differentiating between OOD sets according to their "protection" level of in-distribution sub-manifolds.

Paper

Toward Adversarial Robustness by Diversity in an Ensemble of Specialized Deep Neural Network (CanadianAI 2020): In this paper, we propose an ensemble of diverse specialized CNNs along with a simple voting mechanism which creates a gap between the predictive confidences of adversaries and those of clean samples.

Paper

Towards Dependable Deep Convolutional Neural Networks (CNNs) with Out-distribution Learning (DSML 2018): In this paper, we propose an approach to augment CNNs with out-distribution learning in order to reduce misclassification rate by rejecting adversarial examples.

Paper

Sampling from Complex Networks with High Community Structures (Chaos 2012) In this paper, we propose a novel link-tracing sampling algorithm, based on the concepts from PageRank vectors, which has two phases; (1) Sampling the closest nodes to the initial nodes by approximating personalized PageRank vectors and (2) Jumping to a new community by using PageRank vectors and unknown neighbors.

Paper