Question on Experimental Setting for Ablation Study (Table 5) #18
Description
I have a question regarding the experimental setting used for Table 5 in the paper.
It is mentioned that the default training setting was used for memory lengths 2 to 6. Could you please clarify the following:
Was a single model trained with memory length 2 for training and 3 for testing, and then for Table 5, the memory length was varied from 2 to 6 during evaluation using the same trained weights?
Or were separate models trained specifically for each memory length?
A more detailed explanation of the experimental setup would be greatly appreciated. Thank you!