Some questions about the paper & code

[qingzhenduyu]

Hi,
In your paper, you mentioned "The mean and standard deviation (stddev) of the accuracy reported are obtained by repeating the experiments five times.". I wonder to know it is repeated in the same random seed for 5 times or or you choose 5 different random seeds to repeat the experiments. As far as I know, it should be done in 5 different random seed instead of same seed for 5 times.
And I would like to know the the effect of grad_a in tsslbp.py? For warm @up? @stonezwr
Thanks.

[qingzhenduyu]

Another question, in your paper,you mentioned that you use [0, 1, 0, 1, 1] as desired output spike trains,but in your code, the desired_spikes is setted [0, 1, 1, 1, 1]
is there any difference? Or you just made a slip of pen in your paper?

[chenjiefighting]

About the warm up, you can see #2 and about the desired_spikes, you can see #4

[qingzhenduyu]

About the warm up, you can see #2 and about the desired_spikes, you can see #4

thanks for your reply. About the desired_spikes, I have tried some different patterns, different patterns lead to large different performance. And in your paper, you used Van Rossum distance, and I wonder if you have done ablation study on this "Van Rossum distance"?
Hope for your reply.

[qingzhenduyu]

In global_v.py, line 16, "syn_a /= tau_s", and in tsslbp.py, line 61 "partial_a = glv.syn_a/(-tau_s)", and line 78 "partial_a_partial_u*grad_delta[..., t:time_end]*tau_s"
In the above code, I wonder if you divide tau_s one more time uselessly, for you only need to divide tau_s for once instead of divide twice and multiply it.