Following your videos will make sure person will be excellent at ML and DL architecture..... You are pin pointing some of the issues which are not addresed by others... excellent work.. thanks... best wishes
You can totally replace scaling with Batch Normalization. If you add Batch Normalization layer as a first layer in your network, then your inputs will be normalized a.k.a scaled right before being fed into neural network.
Even if Batch Normalization is added as the first layer in the neural network, it is still recommended to normalize the input data during preprocessing. The reason for this is that Batch Normalization is designed to normalize the activations of each layer in the network, but it does not normalize the input data.
i have a question , i have a lager dataset contains 2381 features (ember dataset) when i want to use it and converted to an image grayscale 48*48 , i must remove 77 features but I want to remove the unuseful features how i van remove these 77 features pleas
You just extract the single convolutional layer and make a new model with just that layer. Then supply an input image and visualize the responses. This is the easiest way I can think of. I will record a video on this topic soon... may be in a week.
Thanks for the great video. Do you have results also by applying dropout after batch normalization and would the order of applying {dropout and batch normalization} matter? I am more interested also in MLP case. Thanks!
Sir, why do I get error when I was running BatchNormalization. I look for the solution but I did not find it. So Can you help me? ImportError: cannot import name 'BatchNormalization' from 'keras.layers.normalization' (/usr/local/lib/python3.8/dist-packages/keras/layers/normalization/__init__.py). Is there the other file for BatchNormalization?or it is a function? method? I installed all, Keras, TensorFlow, batch, and normalization. I have a confusing. Thanks before. Greeting.
@@DigitalSreeni I have a question. so i trained a model (typical doc/cat). I am using the saved model file to see how it performs when I modify the weights/bias and test the arrays at various standard deviation values (from .005-.01) during inference. for example, I use a for loop to run the program at .005 sd for 50 times, then I save the number of correct images that came out each time. I do this for each sd value. I then print out the graph of what it looks like. The problem is that I think I should get higher number of correct images closer to .005 and lower correct number of images the farther the sd gets. but that's not what happens. i get different results each time I run the program. is this expected/normal? sorry i hope this makes sense. this is part of my senior capstone project and I have no idea what is causing this behavior.
hi if you kindly put another link to yann.lecan.com/exdb/punlis/pdf/lecun-98b.pdf as this link transferes you to a chinese with chinese language nothing to see or downlaod thank you
You are the best one that explains the deep learning concepts .. thank you very much dear teacher
Following your videos will make sure person will be excellent at ML and DL architecture..... You are pin pointing some of the issues which are not addresed by others... excellent work.. thanks... best wishes
Awesome explanation !!!! Keep sharing your valuable knowledge
Thank you so much for your effort. It helped me a lot.
Awesome! Thanks for sharing your knowledge. It's very informative.
Wonderful tutorial. Many thanks, Sir
Thanks a lot for your videos. :)
Thank you so much!
You're welcome!
thank you sir. im new into ur channel 💙
You can totally replace scaling with Batch Normalization. If you add Batch Normalization layer as a first layer in your network, then your inputs will be normalized a.k.a scaled right before being fed into neural network.
Even if Batch Normalization is added as the first layer in the neural network, it is still recommended to normalize the input data during preprocessing. The reason for this is that Batch Normalization is designed to normalize the activations of each layer in the network, but it does not normalize the input data.
Nice lecture but why the dropout is used before the max pooling later. ?
How do you scale inputs when it comes to using a trained model?
i have a question , i have a lager dataset contains 2381 features (ember dataset) when i want to use it and converted to an image grayscale 48*48 , i must remove 77 features but I want to remove the unuseful features how i van remove these 77 features pleas
Thanks
First, thank you so much.
Second, I would like to ask, how can I visualize the output of Conv layer of the trained model?
Uhh...can you tell me what is the output of your training network??
@@surajshah4317 For example, the last output of the CNN is a label class, but I need to visualize the output of Conv layer
You just extract the single convolutional layer and make a new model with just that layer. Then supply an input image and visualize the responses. This is the easiest way I can think of. I will record a video on this topic soon... may be in a week.
@@DigitalSreeni Looking forward to your video
Normalizing is ok but while using the trained model we need to normalize the input with the same scaling parameters. How do we do that?.
Thanks for sharing!!Can we use dropout in time series forecasting with deep learning methods ?
Yes, you can
When u finish the explanation of this topics could you make some videos about VAEs? 😬
I'll try. Thanks for the suggestion.
Thanks for the great video. Do you have results also by applying dropout after batch normalization and would the order of applying {dropout and batch normalization} matter? I am more interested also in MLP case. Thanks!
Sir, why do I get error when I was running BatchNormalization. I look for the solution but I did not find it. So Can you help me?
ImportError: cannot import name 'BatchNormalization' from 'keras.layers.normalization' (/usr/local/lib/python3.8/dist-packages/keras/layers/normalization/__init__.py).
Is there the other file for BatchNormalization?or it is a function? method? I installed all, Keras, TensorFlow, batch, and normalization. I have a confusing. Thanks before. Greeting.
If you are trying to perform batch normalization, try keras.layers.BatchNormalization
@@DigitalSreeni Thank you sir. I'll try it.
hey i have the same lamp haha
You seem to be a wise man with good taste in lamps 😌
@@DigitalSreeni I have a question. so i trained a model (typical doc/cat). I am using the saved model file to see how it performs when I modify the weights/bias and test the arrays at various standard deviation values (from .005-.01) during inference. for example, I use a for loop to run the program at .005 sd for 50 times, then I save the number of correct images that came out each time. I do this for each sd value. I then print out the graph of what it looks like. The problem is that I think I should get higher number of correct images closer to .005 and lower correct number of images the farther the sd gets. but that's not what happens. i get different results each time I run the program. is this expected/normal? sorry i hope this makes sense.
this is part of my senior capstone project and I have no idea what is causing this behavior.
hi if you kindly put another link to yann.lecan.com/exdb/punlis/pdf/lecun-98b.pdf as this link transferes you to a chinese with chinese language nothing to see or downlaod thank you
its completely wrong ur hiperlink