Model-Driven Analysis

Model development is can be incredibly hard. From planning models to logical theories, many of the bugs are subtle and hard to detect. This project aims to supercharge a symbolic modeller’s ability to debug and diagnose their models, and will do so through encodings of the analysis problem directly as a model.

(image generated using DALL-E 2 with the prompt, “Understanding a complex symbolic model.”.)

Project Summary

The creation of symbolic models is fraught with subtle bugs that are hard to detect. Unlike commonly used programming languages, debugging techniques are rarely available for declarative languages in the field of Artificial Intelligence. This project aims to remedy that by taking a principled focus on how we might analyze these models through new auxiliary encoding, where solutions to the new encodings yield interesting insight about the model, and how/why it may be failing. The outcome of the work has the potential to be a powerful tool for researchers and practitioners alike.

While this technique may be applicable to several symbolic AI settings, the student will primarily focus on just one of the research areas below. Knowledge in them is an asset, but not required for this position.

Potential Research Areas

Eligibility

This position is (unfortunately) only open to Canadian citizens and permanent residents. Please see the general Queen’s details below as to the reasoning.

MuLab Application Procedure

Expertise in this specific area is not essential, but a passion/interest for exploring and discovering new ideas in the field is a must. Unless explicitly stated in a position posted, the focus of the research must follow closely to what is proposed.

To apply to the above position, email Prof. Muise with the following details:

  1. Your name and a bit of background about yourself. Please include any exceptional circumstances in your career so far (e.g., time taken for health of family reasons, extended industry experience, etc). We’d like to know who you are and we recognize that many successful students have taken unconventional paths in getting here. Help us understand your path and achievements in context of your particular experiences.

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

  3. A CV/Resume that includes any academic/outreach* experience.

  4. Academic transcripts from your past university experience.

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

Once we receive the above, we will send along details on the next steps of the process. Currently, this will include the following:

  1. Reviewing a paper (your choice from a set of 5 by the lab) to provide feedback on.

  2. A coding exercise (roughly) tailored to the position.

  3. One or more interviews with Prof. Muise and students of the Mu Lab.

The Mu Lab typically receives far more interest than we can possibly support. To help demonstrate your interest and ability, below are some suggestions. While none of these are mandatory, they go a long way to establishing your suitability with the lab and help you stand out as a candidate.

  • Being able to read, understand, and reproduce existing research is an important component to conducting your own research. Take on a reproducibility challenge from any of the modern AI conferences (AAAI, IJCAI, ICAPS, ICML, ICLR, NeurIPS, …) from the last 2 years. More information on the process can be found here. The expectation is that you choose something from a paper that would take a weekend or two to reproduce.

  • (for planning-related positions) Much of our research involves planning models (learning them, solving them, and ultimately helping machines to understand them). Concepts for planning are typically taught in most undergraduate AI courses. If you need a refresher, this is a decent tutorial. Model one of the final two domains here using the online editor (you can import a wide variety of existing domains for working examples). To share it, use Session > Save, and send the read-only link.

  • (for PhD positions) Submit a 1-2 page research statement that describes a potential research agenda that aligns with the subject area above.

It’s worth noting that while item (2) is fairly specific to labs such as ours, items (1)+(3) are useful for many other applications you may be considering.

*: The Mu Lab is dedicated to building and fostering a diverse team of researchers. We openly welcome applicants from all walks of life and ask that you emphasize in your application your commitment and any past experience to initiatives that aim to support underrepresented minorities. Such aspects are valued and promoted by members of the Mu Lab, and will be viewed favourably during our selection process.

Queen’s Computing Application Procedure

Every university and department does admissions a little differently. At Queen’s Computing, you can find the official procedure here. There are some additional tips worth mentioning:

  • Unlike some programs that admit students in general without a lab assigned, Queen’s Computing admits students to a specific lab. It is extremely rare to have your application considered unless you have a faculty “championing” your file. Strong domestic students are usually considered with or without faculty support, but in general you must have a faculty member willing to bring you into the lab for your application to be reviewed.

  • For Canadian Universities, the federal goverment provides incentives for the domestic students admitted (i.e., Canadian Citizens and Permanent Residents). At Queen’s, this translates to a large discrepency in the funding required for a domestic -vs- international master’s student. This is why you may find some positions that are available only for domestic students (particularly for newer PI’s with limited funding). Recently, funding for international PhD students has been brought in line with the domestic funding.

  • The Mu Lab is committed to recruiting a diverse candidate pool for every position we hire. While it is not yet a departmental or university policy, applicants that are invited to interview with the Mu Lab and are from (citizenship & corresponding address) developing countries will have their application fee waived/refunded. The complete list of countries that qualify can be found here.

Resources

Applying to grad school can be tough and confusing. The details above will help shed some light on how it works with Queen’s and Mu Lab in particular, but we (as a lab) have also identified some things that can help in general.