Structure

The format of the paper battles will be as follows: 

• Two groups of students will give a brief 20min overview of their assigned paper (2 papers in total).

• Following both presentations, each group will present slides with 3 reasons why their presented paper is better than the paper presented by the other group. You can put anything in these slides that would strengthen your arguments to the audience.

• Afterward, we will have a brief discussion allowing each group to rebut another group's points.

• Lastly, the students in the class who were not presenting will vote on which paper is better (and provide a justification for their vote).

• The winning group will receive prizes.

Suggestions for Arguing for Your Paper

• You should first identify the main strengths of your paper and the main weaknesses of the paper that the other group is presenting. While presenting the arguments for your paper, you should point out the weaknesses of the other paper and try to connect them to the strengths of your paper, i.e., how your paper addresses those weaknesses.

• While arguing for your paper, your main focus should be on the impact of the paper, i.e., convince your audience that your paper had a significant contribution to the research community. Below are some of the things to pay attention to while trying to assess the impact of the paper.

• Technical contribution. Does the paper introduce a fundamentally different framework to a given problem or does it heavily borrow/build on prior work and introduces only incremental changes?

• Simplicity. Is the proposed paper easy to adapt and build on for other researchers? Or is it cumbersome and difficult to reproduce? Do the authors make their code publicly available and how widely adopted their code is? Simplicity should be valued as long as the proposed approach is effective.

• Empirical Results. Does the proposed method in your paper achieve better results than the competing methods? While the empirical results are important, they are often relied on too heavily. Just because one approach achieves better results, doesn't mean that it's better, i.e., maybe better results are achieved while using more computational resources, training on more data, using a significantly more complex system, etc. All of these factors should be taken into consideration.

• Fair Comparisons. Related to the point above, you should pay attention to whether the two approaches can be compared fairly, i.e., are they trained under exactly the same settings (e.g., optimization schedules, pretraining data, computational budget, etc.).

• Presentation Quality. Is the paper easy to read and understand? This is also an important consideration, which can significantly affect how widely the paper will be adopted, i.e., if the paper is difficult to understand it's unlikely that many researchers will adopt it for their own research.