* Proposed and implemented a multimodal deep learning system to identify deception by using feature representations learned from multimodal sets (video and audio) through Restricted Boltzman Machine (RBM).
* By using this multimodal method, the accuray of deception detection on verbal only court can be improved by 300%, compared to the traditional unimodal method.
* Implemented an interactive narrative with a fully animated user-controllable player using parametric behavior tree.
* Created several novel affordances using inverse kinematics and a controllable isometric camera.
Game Demo: [Youtube Link 
]
* Implemented the semi-supervised DDI-LDA model based on Bayesian modal complemented with knowledge-driven distant supervision, instead of the traditional supervised SVM model, to identify the DDIs in biomedical text.
* Applied one filtering process, which utilizes the machine-learning approach of Hidden Markov Models (HMMs), making our DDI-LDA approach more robust to unbalanced data (Accuracy of HMMs is 96.44% over 6,976 datasets.
* Calibrated the camera of a robot vehicle using SVD and Linear Least Squares methods
* Implemented camera calibration from multiple images of 2D planes and augmented these photos with virtual objects (e.g., mapping clip art images and 3D objects onto the photos.)
* Created a MATLAB program to synthesize a large scale image from sample textures.
* Implemented object removal and region filling, which can be widely used for image reconstruction and retouching.