Objectives¶
The team designs and runs experiments in which the weaknesses found in Critical Analysis become measurable facts, and then compresses the whole critique, theoretical and empirical, into the single poster that the poster workshop grades.
Designing the demonstration experiments¶
Pick the families that fit the weaknesses on the table. The menu below suggests the shapes that usually land:
Assumption-violating data. Build or pick inputs that break the assumptions the critique found, and measure what the method’s quality does as the assumption bends. A degradation curve in front of a stated assumption says more than pages of prose.
Adversarial and out-of-distribution settings. Push the method past the regimes its evaluation covered, and show where its behavior changes.
Component ablations. Remove or degrade the pieces the theory credits the method’s properties to, and watch which properties survive.
Replacement baselines. Swap a credited component for a plainer alternative and check whether the theory’s claim of necessity holds up.
Scaling runs. Confront the complexity claims with wall-clock reality at sizes the theory says something about.
Sensitivity sweeps. Vary one configuration knob at a time to expose how brittle the method is to a factor the paper treated as settled.
The implementation and tracking rules of Experimental Reproduction hold here in full: reimplement from the description, fixed documented seeds, runs on the platform the team chooses, every run tracked. Use the same public W&B project and tag the runs of this lab separately, so the two labs stay distinguishable at a glance.
The poster¶
The single page is the discipline: the poster must fit on one portrait A2 page, and it carries the issues that matter with the evidence that shows them. The deck already told the full story. The poster is the distilled argument, and the workshop is where the team defends it in the open.