Capstone Undergraduate Projects

Overview

Level Project Scope Available
499 Coffee/Office Analytics 2-3 (3)
499 LOCM Planning Model Learning 2-3 (1) 🔵
499 Planning.Domains API 2-3 (2) 🔵
499 Nurse and Physician Scheduling 2-3 ✔️
499 Computational Art Installation 3 (3)
500 Planning Graph Factorization 1
500 Dialogue Agent Believability 1

Note: The 🔵 symbol means some students have already signed up (# indicated in brackets), but spots remain.

Capacity

We won’t be able to fill every project position. This is a rough measure of how much capacity Prof. Muise and the MuLab has remaining for capstone projects:


Application Procedure

To apply for any of the projects, please email Prof. Muise with the following details:

  1. Your name and a bit of background about yourself.

  2. The project name and expression of your interest in the area.

  3. Your Queen’s transcript.

  4. (if available) A CV/Resume

  5. (if available) A link to any software/projects you’ve worked on (e.g., GitHub profile).

Over the coming months, we will reach out to interested students and possibly interview if there is high demand for the project. If necessary, the interview process will involve a small coding exercise as well as meeting with Prof. Muise and/or Mu Lab members.


Project Ideas

A short summary of each project is below.

Coffee/Office Analytics

As a collaboration between the Mu Lab and the Machine Intelligence & Biocomputing (MIB) Lab, we have embarked on a research exploration to optimize several aspects of the office experience. This project explores the challenges and opportunities presented by the IoT elements in the MuLab. These include:

(1) Improving the analytics acquisition (website to acquire coffee/tea consumption data).

(2) Running machine learning algorithms to predict the coffee/tea selection given energy profiles (from smart plugs).

(3) Predicting the consumption levels and timing based on timing factors.

More sub-project ideas may be added over time.

(image generated using DALL-E 2)


LOCM Planning Model Learning

This project will involve reproducing some key results in the field of learning for automated planning. Building on the MACQ project, the goal is to reproduce a series of results that can convert sequential data (in the form of actions that have occurred) to a planning model; complete with action specification, world description, etc. This is a very specific form of learning, and the aim is to reproduce the 3 papers related to the LOCM system.

Planning.Domains API

api.planning.domains aims to be the most complete compendium of benchmark problems for the area of AI known as Automated Planning. It currently houses thousands of the existing problems that have been created over the last several decades, and this capstone project aims to dramatically expand its coverage to (1) several other forms of planning; (2) several of the newer benchmarks that have been released since the service was created; and (3) a range of custom features that could be used for machine-learning settings in the field.

The skills that will be used as part of the project include (a) general understanding of automated planning (e.g., what would be acquired from CISC 352); (b) web development (e.g., node.js, flask, etc); and (c) large-scale compute experiments (e.g., using ansible). Please note that none of these are requirements. Rather, these are skills useful to have and will otherwise be a part of the learning process for this project.

The team will be working with Prof Muise and several other researchers internationally to expand on the api.planning.domains service, and the work will impact hundreds of researchers globally that use the service on a regular basis.

(image generated using DALL-E 2)


Nurse and Physician Scheduling

An optimized schedule can make all the difference between a hospital wing that runs smoothly or a hospital wing that is unbalanced. This project aims to explore the problem of scheduling nurses and physicians on one or more hospital floors, and do so using state-of-the-art constraint-based AI optimization techniques. The team will work with a company that specializes in nurse/physician scheduling to generate realistic synthetic data and model complex real-world constraints. The primary software to be explored is MiniZinc, but previous knowledge of the language is not required. Ideal candidates will be those with CISC 352 experience (performing well in the CSP assignment) and those that have done a course project in CISC 204.

(image used with permission from Mesh AI)


Computational Art Installation

This project will involve working with some number of members from the MuLab to visualize, artistically, some of the ongoing research. The final artefact will be an installation in Goodwin 627 (the MuLab), and this open-ended project is aimed at students who are in COCA and interested in the field of computational art.

(image generated using DALL-E 2)


Planning Graph Factorization

In viewing state transition systems as graphs, we are able to ask questions concerning their representation. Does the formulation hide structure that can be exploited to solve the problem? Can we extract any properties or missing information from the formulation? Given some representation of a planning problem, can we determine an equivalent representation that is more amenable to existing techniques? Such questions suggest a novel area of research focused on the representation and reformulation of planning problems and these are the core problems being to be addressed in this project.

(image generated using DALL-E 2)


Dialogue Agent Believability

This project is aimed at understanding conversations between humans and structured dialogue agents. Specifically, it involves understanding how likely different dialogue paths are and identifying unlikely outcomes in dialogue agent interactions. This understanding could allow us to identify edge cases in a dialogue tree and whether they are covered or not by the agent. This would allow improving existing dialogue agents, which would improve the bandwidth of human computer interaction.

(image generated using DALL-E 2)