CS81 Final Project

• You may explore any of the systems that we discussed this semester:
  • Training neural networks with back-propagation
  • Evolving neural networks with NEAT, real-time NEAT, or hyper NEAT
  • Novelty search
  • Intrinsic motivation methods, such as IAC or CBIM
  • Unsupervised learning methods, such as growing neural gas
  • Imitation learning methods, such as HAMMER
  • Reinforcement learning methods, such as the Horde
  • Developmental encodings, such as CPPN
• Expand on and improve your midterm project.
• Try to reproduce an experiment we read about.
• Devise a new experiment using a system we discussed.
• Try to improve on a system we discussed.
• Invent your own adaptive approach using aspects of any of the systems we've discussed.
• Explore an adaptive technique that we did not discuss in class.

Your project proposal should be at least two pages long. It should include a clear description of:

• The adaptive method you will be testing.
  Will you be using existing software? If so, you should test it now and ensure that you understand how to use and modify it for your purpose. If you are writing software from scratch, describe your implementation plan.
• The learning task.
  If you will be exploring a robotic problem, describe the environment, learning goal, and agent that you will use in the experiment. What sensor inputs and motor outputs will be employed? Will you be using a simulated agent or a real robot?
  For all other problems, describe the data that will be used.
• Your hypothesis and expected outcome.
• How you will analyze the results to demonstrate whether your hypothesis was correct.

If you are continuing on the same topic of your midterm project, give a detailed explanation of the extensions you will do for the final project.

Email me a pdf of your proposal by noon on Monday, April 7

Each group or individual doing a project will give a demonstration during the week of April 15. Your experiments do not need to be completed at this point, however be prepared to execute some implemented aspect of your system.

You should plan to speak for 10 minutes with an additional 5 minutes for questions. You are encouraged to use slides for this demonstration. Begin by describing your project (cover the same points as in your project proposal) and then demonstrate some running code.

If you are continuing on the same topic of your midterm project, you must demonstrate a new feature that you have added since completing the midterm.

You will give a 20 minute presentation about your project with an additional 5 minutes for questions. Each group or individual doing a project will be assigned a time to speak during the last week of class.

It may be the case that you are still running experiments at this point. However, you must have some preliminary results to report. If it is a team project, then each team member should speak for an equal portion of the time. You must use slides of some kind.

You should follow these guidelines (put together by Tia Newhall) when creating your talk.

Your paper should be 8-10 pages long and include the following:

• A short abstract of 200 to 300 words summarizing your findings.
• An introduction that contrasts your study with other related work. Find and read at least three articles related to your experiment and discuss these papers here.
• A detailed description of your experiments. There should be enough information provided so that someone could reproduce your experiments.
• An explanation of the results. Use figures and tables where appropriate.
• A discussion of the significance of the results.
• A bibliography.

Your grade will not be based on whether or not your experiment succeeds. Negative results are also useful. Your grade will be based on the design and execution of the experiment as well as the thoroughness and readability of the paper. You do not need to turn in any programs.

You should start writing the paper before the experiments are complete. Begin with the introduction, related work and bibliography. Then move on to the describing the experiments. Only the explanation and significance of the results as well as the abstract need to wait until the experiments are complete.

Email me a pdf version of your paper by 3pm on Friday, May 9.

Your project experiments may require a lot of time to run. Please read the following suggestions for how to conduct your experiments so as not to disrupt the work of other students. Pay special attention to the use of nice and screen.