CS 2400, AI Project

This will cover lab for weeks 8 through 10. You will work in teams to research, implement, and present an AI algorithm that has not been covered in class this term. The goals are

• to research an unfamiliar AI algorithm
• get additional experience in presenting technical content
• locating, tuning, and explaining AI code found online

Details

In pairs (or groups of three where required), you will select a topic from the following list. Each topic may be selected by just one group. You will then find an implementation demonstrating the topic and make a significant change to the implementation. You will then use this experience as the basis for a presentation about your topic. The presentation must include the following:

• An overview of the topic including key researchers in the area
• A clear description of how this leads to AI solutions to problems, especially in modern applications
• A discussion of the strengths and weaknesses of the presented solutions compared to alternative solutions to the same problem
• A demonstration of the system you found that illustrates the solution being presented, including how you modified the system and your experience with those modifications
• Alternative applications for this technology, preferably based on articles you have read (that is, not just how you think it might be useful, but why experts say or think it might be useful in other types of projects)
• An evaluation of the understandability of the system's results such as being able to explain how your system arrived at particular recommendations and how applicable those results are to all populations
• A slide capturing your citations for both the topic and your implementation; you will not present this slide, but it should be available for future readers

Note that a key part of understandability is identifying potential bias in training sets, especially when the training data is leading to a model of human behavior. Common mistakes include training on just college males or completely failing to disclose the makeup of the training population.

Potential Topics

• Bayesian Learning
• Constraint satisfaction systems
• Decision trees
• Dynamic belief networks
• Expert systems
• Hidden Markov models
• Logical agents
• Partial order planning
• Prolog
• Random forests
• Statistical models of natural language processing (NLP)
• Truth maintenance systems
• Word embeddings

Other topics will be considered, but obtain approval from your instructor before spending significant time on the topic. For example, deep neural networks are not on the list because they are covered extensively in later courses.

Demonstration System

As discussed above, you are to find a non-trivial implementation related to your topic, to make a non-trivial modification to it, and to demonstrate a working system to the class. Modifications can be to apply the system to a new domain (one not available online) or to incorporate new features. Simply tuning weights is not adequate, though tuning may play an important part in your solution. You will need to cite where you obtained the implementation. When appropriate, also site key developers.

Further notes:

• It is fine if the project contains a significant amount of code from an online source. Just be sure to document that source and to clearly document your changes or additions.
• "Modification" of the technique can include applying it to a fresh problem. This often includes writing code to prepare data for use, modifying parameters to make the application effective, even spending time researching the domain.
• The focus of the project is not on the modification, but on building an effective presentation that has clear take-aways for the audience. Notice that the presentation is in place of at least one lab; you should plan to spend about as long on it (including preparing code and examples) as you would on a typical lab.

Presentation

It is critical that you review your presentation with your instructor at least two days in advance of presenting the material to the class. Not all team members must be present at this review, but you must leave time to improve the presentation after the review. Presentations will be delivered during the week 10 lab period. All students will be expected to ask at least one question of at least one other team, and the presentation topics will be reflected on the final exam.

Part of your grade will be the effectiveness of your presentation, mostly based on your slide deck. Important notes about slide decks:

• Do not write sentences in the slides! They are too hard for the audience to read.
  • Use fragments and bullets; generally 5-7 words max per bullet
  • No paragraphs!!
• Your basic structure will be to motivate your topic, give a high-level description of it, illustrate it with a demonstration, discuss more detail including advantages and disadvantages, then summarize key points. See above for information that must be mentioned somewhere in the presentation, but you can fit that information in various places.
• It is important to get to demonstrations or examples early; this gives students context to understand the topic more concretely. Waiting until the end for examples means the audience must remember detailed, abstract information until they reach the concrete examples.
• It is tempting to try to get your audience as excited about a topic as you are before demonstrations and details. Avoid that temptation; people will not be convinced of the usefulness of the topic until they see it working and learn more about it. Motivate, but know that what really sells the topic is seeing examples and further discussion.
• While having a slide with references is convenient for documenting your sources on this assignment, do not waste the audience's time going over this slide since no one will read it. Presentations will be published on Canvas for others to retrieve, and they can view the information there if necessary.