Shahrokh Shahi

Coda Building . 756 W Peachtree St NW . 13th Floor . S1320H . Atlanta . GA 30332 . USA
shahi@gatech.edu

Hello there! Every person has their own unique story. Here is a glimpse into mine. My name is Shahrokh /ʃɑ:hroʊk/ but my friends also call me Shah and Sean and I really like, and used to be called, both of them!

I’m a PhD candidate in Computer Science/Computational Science and Engineering (CSE) in the College of Computing at Georgia Institute of Technology, supervised by Elizabeth Cherry. Prior to starting my PhD, I have been a postgrad student at Imperial College London where I earned my second Master’s degree in Computer Science supervised by Alessandra Russo and Mahdi Cheraghchi.

As a person who maintains a life-long passion in computer science and who is also interested in engineering topics and real-world applications of computer science concepts, Computational Science and Engineering (CSE) has been my first and only major to pursue my PhD in, immediately after earning two Master’s degrees, one in Structural Engineering and another in Computer Science.

Here at Georgia Tech, my research interests mainly include, but not limited to Deep Learning, Data Visual and Analytics, and their applications in addressing real-world problems. Particularly, I have been working on developing Interval Deep Neural Networks to account for uncertainty in the input imprecise data. Currently, I’m working on developing robust Reservoir Computing techniques, such as hybrid Deep Echo State Networks, to predict the dynamics of complex systems with partial observability and imprecise data, such as cardiac electrical waves, markets and finance data, etc. Apart from my research, I have also been working as a software developer and enjoying freelance consulting jobs.

Education and Experience

PhD - Georgia Tech 2018 - present

Currently, I am a Ph.D. student in Computational Science and Engineering (CSE) at Georgia Institute of Technology
Major: Computer Science - CSE
Software Developer 2016 - present

After graduating from Imperial College London and before starting my PhD, I worked as a software developer at Alpha Quant company
I have also been working as a full stack web developer
Imperial College London 2015-2016

M.Sc. in Computing Science
Distinction (First Class Honour)
Thesis: Algorithms for Graph Partitioning
University of Tehran 2013 - 2014

Senior Research Assistant,
High Performance Computing Lab
Developing a set of computational packages
University of Tehran 2010 - 2013

M.Sc. in Structural Engineering,
Thesis: Multiscale Simulation of Biomechanical Systems; Human Heart Valves
University of Tehran 2006 - 2010

B.S. in Civil Engineering
NODET 2006 - 2010

NODET (National Organization for Development of Exceptional Talents)
Highschool and Pre-university, major in Mathematics and Physics

Skills

(1) Programming Languages:

(+10 years)

Python

MATLAB

(5 to 10 years)

Java

C++

(1 to 5 years)

Bash

Fortran

(Less than 1 year)

Julia

Perl

Prolog

(2) Machine Learning, Data Visual & Analytics:

(1 to 5 years)

TensorFlow

Scikit-Learn

AWS

(Less than 1 year)

Hadoop

Hive

Spark/Scala

Hbase

SQLite

OpenRefine

Tableau

Gephi

Microsoft Azure

Azure ML Studio

MMap
(PageRank)

(3) Web Development:

(1 to 5 years)

Html

CSS

Javascript

jQuery

(Less than 1 year)

django

Flask

Node.js

Ajax

(4) App Development:

(Less than 1 year)

iOs/Swift

Android

(5) Programming Tools & IDEs:

Git

GitHub

GitLab

BitBucket

Sublime Text

Vim

IntelliJ IDEA

Eclipse

Xcode

(6) Engineering Softwares:

(1 to 5 years)

Abaqus/CAE

SolidWorks

Comsol Multiphysics

HyperMesh

AutoCAD

(7) Others:

Portfolio

The following is a few examples of the projects I have accomplished. There are lots of other projects that I'm going to add them to this collection. So, as long as you can see this text, this section is not complete yet and will be evolving and updating! Click on to see more details.

Feature Selection based on Information Theory

Finding the most informative dimensions (pixels) of face images, given a dataset of two classes of males and females images. The most informative pixels are demonstrated by a heatmap of informativeness scores

Binary Classification: Naive Bayesian Classifier, K-Nearest Neighbors (KNN), and Logistic Regression

Employing three different classification algorithms to classify images into two male and female classes. It is shown that logistic regression works slightly better than the other two approaches

Density Estimation with KDE and GMM

First, running PCA on 13-dimensional "Wine Quality" dataset to reduce the dimension to two and visualize the multi-modal behaviour. Then, using the Expectation-Maximization (EM) algorithm to fit a Gaussian Mixture Model (GMM) to the data and comparing the results with Kernel Density Estimation (KDE) and Histogram

Principal Component Analysis (PCA)

Conducting Principal Component Analysis (PCA) on a dataset of face images to find the principal directions. The analysis shows that by keeping the top 50 dimensions, we can pretty much capture all the information in the dataset.

K-Means Clustering

A Simple K-Means Clustering Test

Traveling Salesman Problem (TSP)

Originally, it was a group project (group of four) in CSE Advance Algorithms course. Even though I was exempetd, I accomplished and submitted this project individually, and won one of the performance competitions. In this project, I developed multiple different algorithms (including Brute-Force, Branch-And-Bound, Approximation, and a variety of Local Search approaches) in Python. At the end, a comprehensive report is presented including algorithms details, empirical evaluations, accuracy tables and evaluation plots (QRTDs, SQDs, etc.).

IFEM Computational Toolbox

An Interval Finite Element Toolbox to Study Uncertainties in Mechanical Structures

IFSA - NAFIPS 2019

Best Student Paper Award

Modeling and Simulation

Cellular Automota, Monte Carlo, Python, Jupyter Notebook

Atlanta House Price Prediction

Machine learning-based interactive price estimation and visualization tool. In this project, I developed the front-end

MapReduce on AWS and Azure HDInsight

Multiple analysis on Google n-grams dataset to obtain statistical information using Apache Pig. At first, the interactive PIG shell has been used to check the procedure step-by-step on a smaller dataset through SSH, then the associated PIG script with all commands has been uploaded on the Amazon Elastic MapReduce (EMR).

In a similar project, a MapReduce program to compute some metrics of a large social media (Friendster) graph has been implemented on Microsoft Azure HDInsight in which I developed a Hadoop code in Java.

Data Visual and Analytics

Using Twitter REST API (through Tweepy library) to retrieve friends and followers information on Twitter. Then, Visualize the obtained network graph using Gephi and calculate the metrics of the graph, such as average node degree, diameter, path length, etc.

Data Visual and Analytics

Using D3 to visualize a force-directed network graph (including visual considerations such as labels, scaling node sizes, etc.)

Data Visual and Analytics

Using D3 to visualize a "wine quality" dataset by creating a set of scatter plots. The dataset provides the quality values based on eight red wine chemical features. The symbol properties, such as sizes and colors, are scaled to improve the visualization.

Data Visual and Analytics

Using D3 to create a collapsible dendogram to visualize hierarchical data. The dataset is a json file.

Data Visual and Analytics

Using D3 to create a heatmap to visualize the number of appearances of characters from each six of the houses in HBO’s Game of Thrones across episodes and seasons.

Data Visual and Analytics

Using D3 to create an interactive bar chart to visualize a dataset decribing the total revenue of all the movies in the given franchise.

Data Visual and Analytics

Using D3 to create a choropleth map of world population and displying a tooltip, on hovering over a country, showing the contry's name, population, etc.

Structural Design Optimization

Developing a MATLAB toolbox implementing optimization approaches for engineering applications.

Structural Design Optimization

Developing a MATLAB toolbox implementing optimization approaches for engineering applications.

Structural Design Optimization

Developing a MATLAB toolbox implementing optimization approaches for engineering applications.

Structural Design Optimization

Developing a MATLAB toolbox implementing optimization approaches for engineering applications.

On-line Learning of User Privacy in Social Networks

Developing a web plugin implementing a Parametric Markov Chain (PMC) model to evaluate the privacy risk and social benefit of sharing information with friends in a social network. In this project, Facebook is considered as the main case study. Accordingly, the Facebook Graph API and related libraries have been employed in the developed Java code to create the back-end.

Particle Simulation on Java

Implementing a Java event driven simulation of the movement of particles undergoing elastic collisions.

Algorithms for Graph Partitioning: Theory and Practice

A systematic study of the major techniques and discoveries in the field of Graph Partitioning algorithms, with an emphasis on the Spectral Graph Theory. Application in clustering and community detection in social networks, and also load balancing in parallel computations. (Master's Thesis in Computer Science)

Designing and Development of a Mobile Robot

Employing robotic kits and Raspberry Pi single board computers to implement robotics concepts, such as wheeled configuration and uncertainty in movement, using sensors to implement reactive behaviors and considering probabilistic localization filters and place recognition concepts,etc. Programming in Python.

MSc Thesis in Structural Engineering

Developing a computational package (MATLAB and Java) for multiscale simulation of biomechanical systems: Human aortic heart valve