Hello researcher:
I want to ask you why you use dropout=0.95 before the output layer. We all know that most people choose dropout=0.5. If it is too large, a lot of important information will be lost. I also tested dropout=0.95 and the results will be very good. I want to know why you use dropout=0.95?
I tested the given code without changing anything on the reported datasets of your paper. I reached only maximal 66% on the benchmark dataset and maximal 54% on the BETA dataset. The only difference is that I trained it on my cpu. Could you help me solve this problem? Because I really want to use it for online robotic manipulation for a master-slave system, due to the results reported in your paper.
0
Thank you for the help earlier. I could reproduce the results of the BETA dataset in matlab, except the chebyfilter. Still gonna work on the filter itself. However, next week I want to start the ssvep datacollection to retrain the model using an OpenBCI headset and use it eventually in online robotic teleoperation. Can I have access to the used datacollection code and the stimulation code created with psychtoolbox? It would greatly help me.
Hi, thanks for your excellent work! When I used your code on the benchmark dataset, I met a problem. When the input signal length is 0.1-0.5s, the training time for the first stage is short( within 30 minutes), but it sharply jumps to 800+ minutes when the input signal length is 0.6s. Is it normal, and why? Has anyone else had the same question?
Thanks!