Intention Prediction in Vehicular Environments


Title: Intention Prediction in Vehicular Environments
Author: Mitch Mathieu
Course: CISC 499

Description: Teaching machines to drive on their own is an ever-evolving task that requires computer algorithms to understand increasingly complex driving situations. Specifically, the ability to predict the near-future actions of other drivers on the road would help prevent our robotic drivers from unnecessary accidents and create a safer driving environment.

For this research project, I used LiDAR and camera data from autonomous vehicles to detect other cars in three-dimensional space and then predict their future positions. The object detection was done using a convolutional neural network trained to detect cars from 3D LiDAR point clouds which produced a 3D bounding box for each car. The future position of each car was predicted by applying a regression model to the current and previous positions of each bounding box.

View Poster (PDF)

Application of Linear Genetic Programming and Metabolomics in Alzheimer’s Disease


Title: Application of Linear Genetic Programming and Metabolomics in Alzheimer’s Disease
Group: Chengyuan Sha
Course: CISC 499

Description:Alzheimer’s disease (AD) is a progressive neurodegenerative disease. There are many people in Canada with this disease because it cannot be diagnosed effectively in the early stage. Metabolomics is the study of low-molecular-weight molecules. These metabolites are a result of complex biological processes, and thus are potential candidates to reflect disease.

Linear Genetic Programming (LGP) is a paradigm of genetic programming that uses a representation of linearly sequenced instructions in automatically generated programs. The term “linear” refers specifically to the structure of the program and does not limit the type of problems that LGP can be used to solve. In this project, I developed a python implementation of LGP with scikit-learn compatible API. I applied this evolutionary learning approach on plasma metabolites data and discovered five key metabolic biomarkers and their interactions. The results demonstrate a disturbance in fatty acid and amino acid metabolism, and the glutamate-glutamine cycle.

View Poster (PDF)

Predicting Tomorrow’s COPD and Asthma ER Visits from Historical Visits and Weather Data


Title: Predicting Tomorrow’s COPD and Asthma ER Visits from Historical Visits and Weather Data
Author: Lixuan Liu
Course: CISC 499

Description: COPD and asthma are respiratory diseases that make breathing difficult and limit the activity levels of individuals. Some of these conditions are made worse by the prevailing weather. Hence, this project aims to investigate the temporal pattern of presentation at ER by patients classified with syndromes asthma and COPD, and to predict tomorrow’s ER visits from previous ER visits data and weather factors for hospitals in Kingston and Windsor. Moreover, ER overcrowding has been seen as a frequent and significant problem. The ability to predict ER visits from prevailing weather factors may provide an opportunity for hospitals to arrange medical resources in advance to ensure that patients will have access to treatments on time and potentially alleviate overcrowding.

View Poster (PDF)

Application of Machine Learning and Metabolomics in Alzheimer’s Disease


Title: Application of machine learning and Metabolomics in Alzheimer’s Disease
Author: Jiaao Chen
Course: CISC 499

Description: Alzheimer’s disease is a neurodegenerative disorder that currently affects half a million Canadians and over 35 million people worldwide. Metabolomic is the study of an entire set of metabolites in a tissue or body fluids like blood or CSF (cerebrospinal fluid) and provides the complete overview of the status of the tissue. Metabolomic is the perfect tool to understand the complex biochemical and biological mechanism of the disease pathology with biomarkers in this case Alzheimer’s disease.

View Poster (PDF)

 

Artistic Line Rendering of 3D Models


Title: Artistic Line Rendering of 3D Models
Author: Lauren D. Bhagwandat
Course: CISC 499

Description: Non-photorealistic rendering (NPR) generates stylized images that attempt to imitate traditional 2D artistry. A particularly challenging area in NPR is producing line rendering of 3D models in order to effectively portray feature contours in a visually appealing manner. This project presents an efficient algorithm for real-time line rendering of 3D models that programmatically achieves varying thickness based on artistic rules often used in illustration.