Sign in

Engineer Project

Clemson, South Carolina, United States
January 23, 2018

Contact this candidate


Md Ashfaqur Rahman

ana Ave *** Orchard St, Apt. 201 Cell: +1-262-***-****

Central, SC, 29630,

Summary: A self-motivated individual seeking a full-time position as a machine learning engineer.

Have a Ph.D. with proficient research experience on recurrent neural networks.

Well versed with machine learning languages and frameworks like Python and TensorFlow.

Experienced in using clusters to solve machine learning problems.

Possess a unique blend of analytical, technical, computational and interpersonal skills.

Education: Doctorate of Philosophy – Electrical Engineering, G.P.A: 3.85/4.0 Aug’14 – May’17

Clemson University, Clemson, SC, USA

Dissertation title: A Distributed Dynamic Estimator Using Cellular Computational Network.

Master of Science – Electrical Engineering, G.P.A: 4.0/4.0 Aug ‘11 - Aug‘12

Texas Tech University, Lubbock, TX, USA

Thesis title: False Data Injection Attacks with Incomplete Information.

Experience: Research Assistant Jan’15 – May’17

Real Time Power and Intelligent Systems (RTPIS) lab,

Clemson University, Clemson, SC

Application of Elman recurrent neural networks on state prediction for power systems

Application of multiple computational intelligence methods on state estimation

Trainee of Data Collection Oct’12 - Dec’12

3S Network Inc., Norcross, GA

Collecting data for radio signal quality assessment

Research Assistant Aug’11 - Aug’12

Texas Tech University, Lubbock, TX

Developed false data injection attack under transmission line admittance uncertainty

Current Project:

Toxic Comment Detection:

A project in Kaggle to detect the toxic comments made on social media. It is an NLP problem that classifies the level of toxicity in six different states as toxic, severe toxic, obscene, threat, insult, and identity hate. The main challenges are to clean the text, and to equate the classes in training data.

Software Published:

“Torrit” – Published the beta version of a power system simulation tool in October, 2017. It is developed with Python tkinter, numpy, jsonpickle etc. It can be downloaded from,


Grocery Demand Prediction:

Participated in a competition in Kaggle to predict the demand of more than 4000 items for 50 different stores. The data were given for around 3 years with different factors that may affect the decisions like price of oil, holidays events etc. The training set contained around 4.8 GB of data that makes the problem a real challenge. I have used Palmetto cluster from Clemson University to run the kernel.

Elman Recurrent Neural Network Based State Prediction

Elman Recurrent Neural Network (ERNN) is a special type of neural network for predicting time-series signals. Due to the feed-back of the hidden layer outputs as the context layer, it requires a special training method called Back-Propagation Through Time (BPTT). In this project, the network is trained to predict the states of a power system. Through rigorous training, the mean absolute percentage error is taken below 1% for system angles. This project is funded by National Science Foundation.

Parallel processing of Constant Jacobian Method

The process of Weighted Least Squares (WLS) takes a long time specially for high dimensional problems like state estimation. In this project, the linear version of the WLS estimator named as constant Jacobian method is parallelized on an NVIDIA GPU (CUDA). To the best of the researchers’ knowledge, this is the fastest estimator ever presented in the literature. This is funded by Department of Enenrgy.

Particle Filter Using CUDA-MPI:

Particle filter is a particle based heuristic algorithm which is very suitable for parallel implementation. In this project, a particle filter algorithm was developed to estimate the states of a power system using CUDA and MPI.

Optical Character Recognition:

In this project, a complete OCR is developed from scratch. It reads from the image of a table, and writes the table in an excel file. It includes the option for designing the characters. The designed characters are compared with the detected characters from the table. The accuracy was around 90%.

Graduate Level Coursework:

High performance computing with GPU

Smart Grid

Computer Networking

Special topics: Neural network

Computer applications in power systems

Software Skills:

Machine Learning Framework: Google’s TensorFlow, Keras

IDE: Jupyter notebook, Anaconda 3

Programming Languages: Java, Python (Numpy, scikit learn)

High Performance Computing: CUDA, MPI

Computing languages: MATLAB, Maple

Database: Pandas, SQL

Related Skills: TensorFlow, Deep learning, Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), attention layer, Convolutional Neural Network (CNNs), supervised learning, unsupervised learning, natural language processing.

Certification: E.I.T. (Engineer in Training)- NCEES Fundamentals of Engineering Exam (FE)


Second in student paper award in Power Systems Conference, 2015, Clemson University

90+ citations in one paper in the last five years

Selected Journal Papers:

1.Rahman, M A, and Venyagamoorthy G K, “A hybrid state estimator for power systems using cellular computational network” Engineering Applications of Artificial Intelligence, Vol. 64, pp 140-151, September, 2017.

2.Rahman, M A, and Venyagamoorthy G K, “Convergence of fast state estimation for power grids” SAIEE Research Journal, Vol. 108, No. 3, pp 117-127, September, 2017.

Selected Conference Papers:

3.Rahman, M A, and Venayagamoorthy G.K. “Power system distributed dynamic state prediction” Computational Intelligence (SSCI), 2016 IEEE Symposium Series on. IEEE, 2016.

4.Rahman, M A, and Venayagamoorthy G.K. “Cellular Computational Generalized Neuron

Network with Cooperative PSO for Power Systems” Neural Networks, 2017 International Joint Conference on. IEEE and INNS, 2017.

Github Profile:

Contact this candidate