Puria Azadi Mohadam

pazadimo {at} sfu {dot} ca

Simon Fraser University

My name is Puria Azadi, a M.Sc. student in Computing Science at Simon Fraser University(SFU), and I am working under supervision of the Prof. Hefeeda. Before joining Simon Fraser University, I was an undergraduate Electrical Engineering (Telecommunication) student with minor in Computer Engineering at the Department of ECE, University of Tehran.

My research interest is at the intersection of multimedia, computational photography, computer vision with strong connections to optimization and machine learning.


CV / LinkedIn / GitHub

Publications

“Revealing True Identity: Detecting Makeup Attacks in Face-based Biometric Systems,” ACM MM, 2020

A. Arab, P. A. Moghadam, M. Hussein, W. Abd-Almageed, and M. Hefeeda

Deep Reinforcement Learning for Dynamic Reliability Aware NFV-Based Service Provisioning,” IEEE Globecom, 2019

H. R. Khezri, P. A. Moghadam, M. K. Farshbafan, V. Shah-Mansouri, H. Kebriaei, and D. Niyato


Education

M. Sc. in Computer Science

Graduate Research Assistant, SFU - Huawei Visual Computing Joint Lab and NSL Lab, School of Computing Science, Simon Fraser University, Canada Advisor: Prof. M. Hefeeda GPA: 4.33/4.33 (A+)

Courses: Statistical Machine Learning, Deep Learning, Computational Photography, Machine Learning, Design Algorithms

B. Sc. in Electrical Engineering

Minor in Computer Engineering

Undergraduate Research Assistant, Advanced Mobile Communication Lab, School of Electrical and Computer Engineering, University of Tehran, Iran Advisor: Prof. V. Shah-Mansouri GPA: 3.72/4. (17.42/20)

Courses: Statistical Inference, Intelligent Systems , Discrete Signal Processing, Signals and Systems, Numerical Computation, Linear Control Systems, Engineering Mathematics, Advanced Programming, Discrete Mathematics, Data Structure and Algorithm, Computer Architecture

Research Experiences

Revealing True Identity: Detecting Makeup Attacks in Face-based Biometric Systems:

NSL Lab at Simon Fraser University and ISI at the University of Southern California

- Face-based authentication systems are among the most commonly used biometric systems, because of the ease of capturing face images at a distance and in non-intrusive way. Makeup attacks are the hardest to detect in such systems because makeup can substantially alter the facial features of a person.

Paper - Code(Available Soon)

More Info

- In our solution, we design a generative adversarial network for removing the makeup from face images while retaining their essential facial features and then compare the face images before and after removing makeup. Also, we collect a large dataset of various types of makeup, especially malicious makeup that can be used to break into remote unattended security systems. This dataset is quite different from existing makeup datasets that mostly focus on cosmetic aspects.

- Our results show that the proposed solution produces high accuracy and substantially outperforms the closest works in the literature.

Improving Visual Question Answering Using Semantic Analysis and Active Learning:

- In this work, we aimed to train a model for the task of visual question answering, using only a small number of labeled data. In order to do so, we proposed a solution influenced by the active learning. We defined an oracle to provide a label for the question that is asked about an image. This oracle is an image captioning network that given an image as its input, generates a sentence describing the objects which are visible in that image.

Report - Poster - Code

More Info

- We used a semantic similarity calculator to connect the result of the image captioning model and interpret that to become a potential label for the visual question answering task. By using this structure and defining a new loss function, we became able to train a visual question answering model using a small number of labeled data.

- The simulations of our proposed method shows that this approach can result in a model, which performs in the same standard of classical approaches.

Deep Reinforcement Learning for Dynamic Reliability Aware NFV-Based Service Provisioning:

Advanced Mobile Communication Lab at University of Tehran

- Network function virtualization (NFV) is referred to the technology in which softwarized network functions virtually run on commodity servers. Such functions are called virtual network functions (VNFs). One of challenges is to meet the reliability requirement of the requested services considering the reliability of the commodity servers.

Paper

More Info

- To address such an issue, in this paper, we employ Deep Reinforcement Learning (Deep-RL) to model NFV placement problem considering the reliability requirement of the services.

- Numerical evaluations show that the introduced model can significantly improve the performance of the network operator.

Notable Projects

Iterative Edge Aware Filtering and Cross Filtering

The implementation of spatial filtering proposed in "Temporally Coherent Local Tone Mapping of HDR Video". The most relevant parts are sections 4.1 and 4.2 of the paper.

Course: Computational Photography and Image Manipulation
Code

Texture Synthesis and Texture Transfer

The implementation of the texture synthesis and texture transfer method proposed in: Efros and Freeman , "Image Quilting for Texture Synthesis and Transfer".

Course: Computational Photography and Image Manipulation
Code

Poisson Image Blending

The implementation of the Poisson image blending method proposed in: Perez et al., "Poisson Image Editing"



Course: Computational Photography and Image Manipulation
Code

Naive Bayes Classifier to Predict Party Affiliation

Implementing and benefiting the NB classifiers to predict the party affiliation of either Democrat or Republican of Congressmen based on their votes for 16 different measures in python.

Course: Statistical Machine Learning
Code

Gibbs Sampling for the Image Restoration

Estimating the posterior probabilities of pixels value and restoring the noisy images by using the Monte Carlo estimate and Gibbs Sampling which iterates and samples in Ising model of the image in python.

Course: Statistical Machine Learning
Code

Factor Graphs and Loopy Belief Propagation

Error correcting codes based on highly sparse, low density parity check (LDPC) matrices, and using the sum-product variant of the loopy belief propagation to estimate partially corrupted message bits in python.

Course: Statistical Machine Learning
Code

Implementing and Analyzing Autoencoders

Implementing, analyzing, and comparing the Fully Connected, Convolutional, and Variational Autoncoders using Keras in python.



Course: Deep Learning
Code

RNN Sequence Processing

Implementing RNN-based language models and compare extracted word representation from different models using Keras in python. Designing Vanilla RNN to capture word representations for classifying the 20Newsgroups dataset.

Implementation and Visualization of CNNs

Design and implementation the LeNet using Keras in python to classify the Caltech101 containing 101 categories. Visualization the network's weights and analyzing them was the final objective.

Course: Deep Learning
Code

Get in touch at: pazadimo {at} sfu {dot} ca