Table 2.
Input noise (%) | Brian (%) | SpiNNaker (%) |
---|---|---|
0 | 94.955 | 94.94 |
60 | 88.665 | 88.66 |
The first column shows results for a software simulation in Brian with Q3.8 precision, the second column shows results for SpiNNaker, which uses the same fixed point representation of weights. Differences in performance are almost negligible.