Questions Merge to Learn - ufal/NPFL095 GitHub Wiki

  1. The paper describes two baseline methods (CF and RT) and three PTM methods (Task Arithmetic, Linear Interpolation, WiSE-FT). What are the advantages and disadvantages of these five methods? When would you use which one?

  2. PTM is reported to be much faster (less training steps needed) than RT. Do you understand why? Is PTM faster than CFT? Why?

  3. Can you annotate at least some points in the subfigures of Figure 1 with their ω values? (You can describe it in words or attach an image, whatever is easier for you.)

  4. Bonus: In the first subfigure of Figure 1 (Science), why are both ends of the line at the bottom of the plot?