Syamil Mohd Razak's Projects
Applied Natural Language Processing graduate course.
A simple guide to a vanilla CNN for classification and transfer learning
A simple guide to a vanilla CNN for regression, potentially useful for engineering applications.
MS Kinect skeletal tracking children's game designed for CS Mott Children's Hospital.
Database management graduate course
2D convolutional autoencoder and variational autoencoder implementation for tutorials.
Tutorials on classic dimensionality reduction techniques using tailored bases (SVD) and generic bases (Fourier and Wavelets).
Keras implementation of dcgan, wgan and wgan-gp with digit-MNIST dataset for tutorials.
Implementation and demo for "Convolutional neural networks (CNN) for feature-based model calibration under uncertain geologic scenarios" by Mohd-Razak and Jafarpour (2020) as published in Computational Geosciences.
Gradient-based history matching with high-fidelity models and reduced latent-space representations.
Implementations of graph search algorithms in Python - including Breadth-First Search (BFS), Depth-First Search (DFS), Uniform-Cost Search (UCS) and A* search (with animations!).
Information Retrieval and Web Search Engines graduate course
Lasso regression for classification and salient feature identification.
Ensemble-based history matching method with latent-space proxy model for nonlinear forward model and non-Gaussian models.
Joint parameterization and direct history-matching method for engineering systems. LSI simultaneously extracts salient spatial and temporal features for direct inversion. Exploration of meaningful latent spaces allows rapid generation of ensemble of relevant inversion solutions.
Latent space XGBoost for regression of latent variables representing high-dimensional multivariate timeseries in unconventional reservoirs.
Python implementation of a minimax AI agent for a game of Go with board dimension of 7x7. The minimax agent has some Go-specific heuristics and can be played with a random agent.
Simple Python implementation of a multilayer perceptron for image classification
Neural networks with embedded Decline Curve Analysis function.
Pangman is a game (a rather amateur one) that blends Hangman and Pacman together! The goal of the player is to guess a word by navigating Pacman to collect correct letters. The player also needs to avoid getting eaten by any ghost or choosing too many incorrect letters.
A brazen two-column theme for Jekyll.
Time-series forecasting with 1D Conv model, RNN (LSTM) model and Transformer model. Comparison of long-term and short-term forecasts using synthetic timeseries. Sequence-to-sequence formulation.