AVoE: A Benchmark on Predicting Violation of Expectation for Artificial Cognition [Pre-print] [NeurIPS Workshop Paper]
I developed a 3D Synthetic Dataset, AVoE, that showcases impossible and possible scenes with heuristics of features and rules, which an artificial agent must label using the Violation-of-Expectation (VOE) paradigm. The scenes are grouped into 5 event categories: barrier, occlusion, containment, collision, support. We also introduced the Object File Physical Reasoning Network (OFPR-Net) which exploits the dataset’s novel heuristics to outperform our baseline and ablation models. The OFPR-Net is also flexible in learning an alternate physical reality, showcasing its ability to learn universal causal relationships in physical reasoning to create systems with better interpretability.
Developed enigma, a software framework API written in Python to simplify the usability and
flexibility of testing and conducting automatic reverse-engineered protocol analysis using multiple
machine learning techniques from various domains.
Introduced a novel unsupervised deep learning approach to automated protocol reverse engineering.
A variety of deep learning architectures were used to generate encoded semantic information of data
packets for clustering into unknown protocols. I also Investigated multiple unsupervised machine learning techniques as baselines for comparison against the deep learning approach. [Code is proprietary and confidential]
A ROS Workspace for the interception of a ball projectile via a simulated drone
Deep RL for SpiderBot   [Report]   [Code]
A custom-designed Spider Robot trained to walk using Deep RL in a PyBullet Simulation
Flapping-Wing MAV Deep RL   [Report]   [Code]
A ROS Workspace for a Deep RL approach to teaching a Physical Flapping-Wing MAV to Fly in a controlled environment
OnitamaAI   [Report]   [Code]
An Artificial Intelligence Learning implementation on the board game Onitama
Machine Vision & Image Processing   [Report]   [Code]
A full workflow of a series of digital image processing tools written from scratch in MATLAB
FrozenLake RL   [Report]   [Code]
A tabular reinforcement learning approach to a custom & classic gridworld setting
Physics Informed Neural Networks   [Report]
A Physics Informed Neural Network predicts the two-dimensional viscous flow field of the flow past a cylinder rotating at a constant rate of rotation in an incompressible flow. The flow is numerically simulated using an open source CFD software tool, OpenFOAM
The PINN model is trained to obey the governing Navier-Stokes equations as a penalising regularisation loss using automatic differentiation in the open source machine learning software tool TensorFlow in which the PINN is developed. The PINN is also shown to have better performance than a neural network (NN) which is not physics based but trained with CFD Data, especially with scarce data. [Code is proprietary and confidential]