Skip to main content
[Preprint]. 2024 Jul 12:arXiv:2407.09100v1. [Version 1]

Table 5:

We used seeds 8, 16, 42, 64, 128, 512, 1024, 2048, 4096, 16384, 32768, 131072, 262144, 104857 to ensemble the factorized benchmark. Here we provide the individual performance of the models on the final test set to analyse how much performance depended on the seed.

Main track
Seed single-trial ρst average ρta
8 0.1932 0.3650
16 0.1642 0.3210
42 0.1887 0.3569
64 0.1780 0.3380
128 0.1839 0.3479
512 0.1799 0.3402
1024 0.1865 0.3528
2048 0.1672 0.3178
4096 0.1734 0.3305
16384 0.1880 0.3571
32768 0.1933 0.3661
131072 0.1852 0.3513
262144 0.1839 0.3488
1048576 0.1943 0.3674
mean 0.1828 0.3472
std 0.0094 0.0159