Term Project

Due date as specified on course syllabus

You are strongly encouraged to form groups of two students for the term project. Project proposals involving a single student can be considered, but the scope of the project should still be considered on par with that of two-student groups.


For the term project, you are to propose an idea that embodies some of the concepts you have learned (or are about to learn) in this AI course. The specific application that you decide to target is left completely to your discretion, but it should be interesting, feasible (within the time frame allotted) and well-suited to the efforts of two students.

The instructor and/or TA will provide feedback regarding your idea, possibly suggesting modifications, expansions or in some cases, narrowing the scope of your proposal. Note that you must obtain approval of the instructor for your proposed project.

Keep in mind that the time frame allocated to the project is similar to that of the previous assignments. As such, the scope of work anticipated for each project member is of a similar nature. Although you are encouraged to explore topics of your own interest, this should not require an investment of effort on par with a Masters thesis! Specifically, you need not work on something brand new; a re-implementation of an existing system is perfectly acceptable, provided that you will learn something and have the opportunity to gain relevant AI experience in the process.

The deliverables will include a report, describing your approach and results as well as an in-class presentation during the final week of the term.

Marking scheme

Introduction & Related Work 0. none or incomprehensible 1. The project's objectives are defined with a brief overview of the problem space. Other work is provided but it is not clear how it relates to the problem space or useful to understanding the project’s chosen approach. 2. The project's objectives are defined and non-trivial. There is sufficient overview of the problem space so readers unfamiliar to the topic have enough background knowledge. A brief explanation of what course concept the project relates to is included. At least 2 related works are clearly described with a clear connection to the project.
Method 0. none or incomprehensible 1. The model implemented and data set used are minimally described. There is little to no explanation of how the method was developed. 3. The model implemented and data set used are described clearly enough that there are enough details that someone could replicate the project. There is an explanation of how the model was implemented and the implementation required more than using open ML libraries (e.g., scikit-learn). There is clear reasoning and justification behind the author's method. 4. Along with the good criteria, there are attempts to improve performance after the first iteration of running the model whether it is by tuning parameters, adjusting training, or changing model(s). There is discussion on the how the performance was affected by these attempts with possible explanations.
Results 0. none or incomprehensible 1. The performance of the model(s) are stated. 3. Graphs or tables are included to show performance of the model(s). The report describes and justifies the measure(s) used for evaluating performance. If more than one model was tested, there is a discussion of tradeoffs between models. 4. Along with good criteria, to better evaluate performance, results of a random baseline and a standard ML method are provided (ML libraries can be used for this). There is discussion on how the authors' model compares.
Discussion & Conclusion 0. none or incomprehensible 1. A clear conclusion is drawn from the results and tied back to the stated objective in the introduction. 2. Along with adequate criteria, the report discusses the successes and failures of their approach based on their results and objectives. Limitation of work and future improvements are discussed
Formatting & References 0. Does not meet format reqs   1. Report is in IEEE conference format report, ideally 4-6 pages in length, including references


The presentation, which accounts for 20% of your project mark, will last approximately fifteen minutes per group (regardless of group size) and should be seen as an opportunity to provide an overview of what you accomplished and the results obtained. Although quality of your oral delivery will not be graded, per se, the articulation of your accomplishments, as evidenced by the presentation, will factor as part of your mark for the overall project. You are thus strongly encouraged to practice your presentation beforehand and make sure that you stick to the time limits. Otherwise, you will be cut off and this may impact your grade. Students are strongly discouraged from alternating back and forth between the two presenters as this is distracting and interrupts a smooth flow.

Marking scheme

relevance to AI 0. Does not embody a concept(s) learned in the course 1. The project embodies a concept(s) learned in class, but does not extend what was learned in class or the project problem is trivial. 2. Project clearly extends a concept learned in class that is moderately challenging enough that the students learned something new outside of the course curriculum.
Clarity of objectives and approach 0. unclear 2. The objectives are defined, but the approach is superficially described. For example, the approach is unclear or not fully justified in relating to the objectives. 3. The objectives and the approach are clearly communicated and supported by a brief summary of previous work done in the domain. There is justification for the approach chosen and how it relates to the objectives. 4. Along with the good criteria, the presentation provides a clear conclusion discussing the pros/cons of their approach based on their results and objectives. Future improvements are proposed.
Technical content 0. unclear 2. The model(s) implemented, the data set used, and final results are minimally described. The results are stated but lacks analysis. 3. Graphs and tables are appropriately used to describe the dataset and performance of the model(s). The evaluation measure for performance is defined and used to present clear results. 4. Along with the good criteria, the group discusses attempts to improve performance. Meaningful analysis is provided on trade-offs faced (e.g. between models, parameter tunings, evaluation functions. This will vary on the project)
Answers to questions 0. poor 1. adequate 2. good

Submitting your assignment

Your submission, made through Moodle, should consist of an IEEE conference format report, ideally 4-6 pages in length, including references. Appendices, if included, to help readers understand your results, are not considered part of the page count. A softcopy of your source code must also be included, with instructions to build and run.

Remember that your report is the main method of communicating what you have accomplished to the reader. Therefore, make sure that it is well organized and well written. Your peers will be reading and assessing this work, so it must be written in a manner accessible to a typical ECSE 526 student.

Last updated on 24 November 2017
by Jeremy Cooperstock