CV

Education

• Postdoc, The Alan Turing Institute, 2022
• Postdoc, Vector Institute, 2021
• Ph.D in Electrical Engineering and Computer Science, Queen Mary University of London, 2021
• M.S. in Electrical Engineering (Digital), Sharif University of Technology, 2016

Work experience

• 2020: Internship
  • Inria
  • Project: Intersection of differential privacy and adversarial audios
  • Supervisors: Aurelien Bellet, Emmanuel Vincent, Marc Tommasi
• 2020: Research Assistant
  • Queen Mary University of London, University of Lisbon, Carnegie Mellon University
  • Project: FoolHD: Fooling speaker identification by Highly imperceptible adversarial Disturbances
  • Goal: Generate and conceal the adversarial perturbation within the original audio files
  • Supervisors: Alberto Abad, Bhiksha Raj, Andrea Cavallaro, Isabel Trancoso
• 2019: PhD Enrichment
  • The Alan Turing Institute
• 2018: Internship
  • The Alan Turing Institute
  • Project: Privacy-aware neural network classification & training
  • Goal: To invent new encrypted methods for neural network training and classification
  • Supervisors: Matt Kusner, Adria Gascon, Varun Kanade

Skills

• Programing
  • Python (PyTorch, Keras, Tensorflow, Caffe2, Caffe)
  • C/C++ (ABY (Multi-Party Computation), SEAL and HElib (Homomorphic Encryption))
  • Matlab
  • Verilog

Publications

A Hybrid Deep Learning Architecture for Privacy-Preserving Mobile Analytic

A New Algorithm for Training Sparse Autoencoders

Privacy-Preserving Deep Inference for Rich User Data on The Cloud

Deep Private-Feature Extraction

Private and Scalable Personal Data Analytics using Hybrid Edge-Cloud Deep Learning

Distributed One-Class Learning

SCENE PRIVACY PROTECTION

QUOTIENT: Two-Party Secure Neural Network Training and Prediction

Exploiting vulnerabilities of deep neural networks for privacy protection

PrivEdge: From Local to Distributed Private Training and Prediction

EdgeFool: An Adversarial Image Enhancement Filter

DarkneTZ: Towards Model Privacy on the Edge using Trusted Execution Environments

ColorFool: Semantic Adversarial Colorization

FoolHD: Fooling speaker identification by Highly imperceptible adversarial Disturbances

Semantically Adversarial Learnable Filters

A Zest of LIME: Towards Architecture-Independent Model Distances

Washing The Unwashable : On The (Im)possibility of Fairwashing Detection

Reconstructing Individual Data Points in Federated Learning Hardened with Differential Privacy and Secure Aggregation

When the curious abandon honesty: Federated learning is not private

Private Multi-Label Classification

Differentially Private Speaker Anonymization

Losing Less: A Loss for Differentially Private Deep Learning

Tubes Among Us: Analog Attack on Automatic Speaker Identification

GAP: Differentially Private Graph Neural Networks with Aggregation Perturbation

Confidential-PROFITT: Confidential PROof of FaIr Training of Trees

Mnemonist: Locating Model Parameters that Memorize Training Examples

Confidential-DPproof : Confidential Proof Of Differentially Private Training

Confidential Guardian: Cryptographically Prohibiting the Abuse of Model Abstention

Nebula: Efficient, Private and Accurate Histogram Estimation

Talks

Teaching

Neural Network

Machine Learning

Lecturer: An Introduction to Trustworthy Machine Learning

An Overview of Differential Privacy, Membership Inference Attacks, and Federated Learning

Collecting Speech and Telemetry Data Privately