PyTorch Tutorial 12 - Activation Functions

[@normalperson1130]

Thanks. This couldn't have come at a better time. I've a project that needs Pytorch in my university and this will be greatly helpful

[@patloeber]

Thanks for watching! I'm glad it is helpful :)

[@alteshaus3149]

Thank you for these great videos. I can really learn a lot and it helps me for my future career

[@Ftur-57-fetr]

Simple, clear, THANKS!!!!!!!!

[@patloeber]

Glad it helped!

[@riadhossainbhuiyan4978]

Great teaching ❤

[@haoranwan1233]

Thanks！The best tutorial I have ever seen!

[@mahmoudabbasi1994]

A great course. Thank you!

[@DanielWeikert]

Thanks for all your effort. Appreciate it! Could you do a video on text generation with pytorch? Best regards

[@patloeber]

Thanks for watching! I have 4 more tutorials planned for this playlist: - feed forward net - convolutional neural net - transfer learning - tensorboard After that I want to do more practical applications, so yeah text generation is a nice suggestion :)

[@beltraomartins]

Thank you!

[@nicolasgabrielsantanaramos291]

Thanks for the class!!!

[@amrutapatil1417]

great tutorial, thankyou so much!!!

[@patloeber]

You're very welcome!

[@HoangNguyen-be4vy]

As far as I'm concerned about your example, this NN has 3 layers: - 1 input layer - 1 hidden layer (uses ReLu function) - 1 output layer (size 1, uses Sigmoid function) Am I understanding it correctly? Correct me if I am wrong.

[@patloeber]

Exactly :)

[@alirezamohseni5045]

thanks, a lot😀😀😀😀

[@VarunKumar-pz5si]

Please do some videos on Computer Vision projects ...................

[@patloeber]

Yes good suggestion!

[@urosgrandovec3409]

without activation functions, a composition of linears transformations is again a linear transformation

[@anonim5052]

great

[@saurrav3801]

Thanks bro I saw it

[@abderrahmanebououden5173]

thanks bro :)

[@joywang8173]

Does anyone knows why tanH is a good option for hidden layers? i'm new to deep learning😊

[@saimakhalil5137]

The tanh function outputs values between -1 and 1, which can help in centering the activations around zero. tanh is also non-linear, which allows neural networks to learn complex relationships in data.Tanh has stronger gradients than sigmoid, particularly around zero. This can facilitate learning and convergence, especially in deeper networks.Unlike sigmoid, which outputs values between 0 and 1 centered around 0.5, tanh outputs values between -1 and 1 centered around 0. This can make optimization easier because weights and biases can be updated in both directions in the weight space.

[@bluebox6307]

If anybody wonders (like i did) if there is any difference between instantiating nn.Sigmoid() or calling torch.sigmoid(), check out the answer from KFrank in discuss.pytorch.org/t/torch-nn-sigmoid-vs-torch-sigmoid/57691/3 :) (and according to discuss.pytorch.org/t/is-there-any-different-between-torch-sigmoid-and-torch-nn-functional-sigmoid/995, torch.nn.functional seems to be deprecated by now)