Proposal
(529 section)Students enrolled for graduate credit are expected to innovate and drive the project themselves. Specifically, they will propose, implement, and characterize at least one solution to parts (3) and (4).
The proposal document is a brief (roughly 3-6 page) narrative document describing a proposed approach to solving these two problems. It should address:
- Proposed idea
- Concisely, but precisely, state the proposed approach to solving the problem. (Mathematical or algorithmic notation may be helpful.)
- Motivation
- Why is this approach expected to work?
- Related work
- What other things have been tried on this problem, and what other problems has your proposed approach been used on? What were the general results in each case, and how does that support your motivation?
- Novelty
- How is what you're proposing new? Note: For a class project, your approach is not required to be really new. You are free to re-use existing techniques, so long as they go beyond the scope of what we have discussed in class. But you should state whether your proposed approach is new to this target problem or is novel in some other way. (You are free to invent new techniques altogether, but not required to.)
- Expected results
- How do you expect your proposed approach to behave/perform? Not just in terms of accuracy, but also with respect to some subset of: Computational performance? (Is this expensive or cheap to do? Numerical stability issues?) Statistical performance? (Statistical stability/convergence, how much data is required, etc.) Form of answer? (E.g., will it produce a classification boundary? A probability distribution? A subset of points from a 3-D cloud? Some other set of features?) Drawbacks/potential failures? (Where could your approach break down, fall apart, or fail?) Not all of these categories will be relevant, and there are other considerations I haven't mentioned here, but it's worth thinking through which of them are and describing as many relevant issues as possible.
- Validation/testing/empirical measurement
- What standards for "success" will you use, and how will you
measure them? How will you assess whether your proposed method is
working, whether it is working better than other possible
approaches, etc? To the extent possible, you should describe specific
experiments you plan to do, including what performance metrics you
will measure, what implementation(s) you will use, etc.
Hint: the more specific and comprehensive you can be here, the better off you will be.
In addition to the above "content" areas, I will also evaluate your proposal on writing clarity and correctness.
