Cifar-10 accuracy ~70% after 300 epochs

[gsrivas4]

Hi,

I ran your code for Cifar-10 with 256 filters on each conv layer. I am getting ~70% accuracy after 300 epochs. I saw in a presentation that you achieved 86.05% accuracy on Cifar-10 with this model. Do I need to change some parameters to get the accuracy value.

Thanks

[BertMoons]

Hm , I'm looking into it. I've pushed a slightly modified version and am currently running the training again. I've cross-verified with the exact code we used for the chip and don't see any real differences. I'll get back to you once I know more. Best 2018-03-17 19:36 GMT+01:00 gsrivas4 <[email protected]>:

…

Hi, I ran your code for Cifar-10 with 256 filters on each conv layer. I am getting ~70% accuracy after 300 epochs. I saw in a presentation that you achieved 86.05% accuracy on Cifar-10 with this model. Do I need to change some parameters to get the accuracy value. Thanks — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AF4nfk2bxFJ--Yzii93qkwyNHBH42PTZks5tfVepgaJpZM4Su6ib> .

[BertMoons]

Hm, I get to >70% in about 12 epochs... My guess is something is wrong with your setup. Anyway I've pushed a version that should work with the .py -f 256 flag out of the box. It might pay off trying out that version.. Good luck!\ Bert 2018-03-17 19:36 GMT+01:00 gsrivas4 <[email protected]>:

…

Hi, I ran your code for Cifar-10 with 256 filters on each conv layer. I am getting ~70% accuracy after 300 epochs. I saw in a presentation that you achieved 86.05% accuracy on Cifar-10 with this model. Do I need to change some parameters to get the accuracy value. Thanks — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AF4nfk2bxFJ--Yzii93qkwyNHBH42PTZks5tfVepgaJpZM4Su6ib> .

[gsrivas4]

I ran the new commit and could get the claimed accuracy value. Thank You.

Thanks,
Gaurav