I noticed that the current benchmarks primarily demonstrate the effectiveness of the method at lower sampling ratios (e.g., 10% - 40%), which is very impressive. I was just wondering if you have also considered testing the model at higher sampling rates, such as 0.5? It would be very insightful to see if the performance gain remains significant or begins to saturate as the sampling rate increases.
I noticed that the current benchmarks primarily demonstrate the effectiveness of the method at lower sampling ratios (e.g., 10% - 40%), which is very impressive. I was just wondering if you have also considered testing the model at higher sampling rates, such as 0.5? It would be very insightful to see if the performance gain remains significant or begins to saturate as the sampling rate increases.