This course counts as “implementation-intensive” in the breadth area “AI”.
This class used to be listed under CS980.
This course will be managed on mycourses. Online discussions on Piazza (as linked on mycourses)
For mutual expectations and details about grading see the Syllabus.
Prologue
Part 1
Part 2
Epilogue
It is important that you don’t just passively read a research paper and blindly believe what’s in it. You will not understand most papers on the first attempt. If you feel lost, no problem, but you need to list all questions/blockers you encountered. There are a few methods to combat this, that I highly recommend one of the following two methods:
The “very smart friend” method: http://www.theexclusive.org/2017/11/read-a-paper.html
The three-pass method: http://ccr.sigcomm.org/online/files/p83-keshavA.pdf (At the very least, you must complete “the first pass” and “second pass” using the three-pass method.)
Both methods recommend that “You should be able to summarize the main thrust of the paper, with supporting evidence, to someone else”.
If you are reading a survey paper, it will discuss a number of different methods. Please select three methods that you will details in your research notes (you should be able to summarize the main ideas).
When you submit “normal” reading notes, you can keep it short and down to the point. One or two paragraphs in your own words (!) are sufficient. Please comment on how this paper applies to our shared task (“it does not” is the wrong answer).
Always comment on task, approach, eval paradigm, and results.
More detailed information about reading notes
You are asked to submit a literature survey for your expert topics. Please select additionally three papers from the “additional reading list” and write a 1-2 paragraph summary for each. Then conclude the survey with a final paragraph that describes how these papers relate to one another. (What do they share? where do they differ?). If you have ideas for what they are missing, please include this as well.
The goal of the prototype is to implement some algorithms from papers on the reading list. This requires to extract the pseudocode from the research paper, as well as many technical details on how to run it (especially when these are based on machine learning). I encourage you to write a technical excerpt before hitting the keys.
It is possible that important details are missing. I encourage you to contact the authors and ask for clarification (“I am a student from University of New Hampshire working with Laura Dietz and I have a question about your paper XXX…”). Probably you will only re-implement certain components of it.
You will have to adapt the setup, for it to apply to the TREC CAR task. Please note preconditions and necessary data transformation to your technical excerpt. I suggest you include these notes in the report you will submit along with the prototype.