3 votes
Accepted

Is it bad to average several MAEs calculated from chunks of a big test dataset?

Can I save the MAE from each chunk of data and then average them ? Yes. This is perfectly fine. Why? Think about the metric's definition. Caveat: We assume $k$ chunks of equal chunk size $cs$. If ...
J_H's user avatar
  • 802
2 votes
Accepted

Train CNN weights by using FFT - Reinforcement Learning?

If you're doing this strictly for learning the inner machinery of how a CNN works, then whipping up something in C++ or python or your language of choice is fine, and can be a good learning exercise. ...
brewmaster321's user avatar
2 votes

How to Findout if a neural network is invariant

What kind of neural network are you using? It could be invariant if the initial weights are always the same, but some NN can also have stochastic inner mechanisms (ex, noise and minibatches). In all ...
Nicolas Martin's user avatar
2 votes

How to correct ValueError about incompatible dimensions of training data set in locallyconnected1D layer NN

To elaborate on Ubikuity's comment, there are a couple of issues. The error that you are seeing is because your model output shape is (None, 98, 10), so it will output a 2d tensor. The y_train shape ...
brewmaster321's user avatar
2 votes
Accepted

Pytorch mat1 and mat2 shapes cannot be multiplied

In forward, the image first passes through some convolutional layers (i.e. self.classifier), then it is flattened, then passed ...
noe's user avatar
  • 25.7k
1 vote

How to optimize transposed convolution?

I suggest using a bilinear interpolation followed by a convolution instead of deconvolution. Deconvolution is prone to checkerboard artifacts. It may also helps in terms of execution speed.
Lelouch's user avatar
  • 151
1 vote

What does it mean if a neural networks starts overfitting more after applying regularisation techniques

The results seems alright. Your training accuracy could be 99% if trained on enough epochs but it does not mean it is a real indicator on how well it will do on unseen data. Regularization bridges ...
Rathod's user avatar
  • 71
1 vote

Why the training accuracy stays high but validation accuracy does not change?

You are using medical images for which the difference in images might not be significant. For this, you can use residual networks or ResNet architecture which has residual connections which improves ...
Rathod's user avatar
  • 71
1 vote

Is improving a Neural Network really just "trial and error"?

Some activation functions work better in some cases. Hidden layers with ReLU work surprisingly well. I would usually recommend adding some regularization to generalize better. Use a variable learning ...
Yash Mali's user avatar
1 vote

Is improving a Neural Network really just "trial and error"?

I'm new to this field as well, and from what I've learnt, yes, a lot of what we're doing while working with NNs is empirical. We mostly build such NNs to work with some specific type of data, and ...
code's user avatar
  • 11
1 vote

Training ResNet50 model for binary classification

Static learning rate is usually a bad thing, and 0.1/0.08 is usually a very high learning rate after a few epoch (for some networks even 0.001 is too high). Easiest thing to do is going with Adam ...
Lelouch's user avatar
  • 151
1 vote
Accepted

What are the general rules or principles for finding matrix operations that are used as filters in convolutional neural networks?

In general, the filters of a convolutional network are NEVER designed, they are trained. Of course, there can be research lines that study the inclusion of hand-designed filters in CNNs, but in the ...
noe's user avatar
  • 25.7k
1 vote

How to align the description of a convolutional neural network in keras with wikipedia's conceptual model?

There are two misunderstandings: How Convolution with layers works How the pooling is defined Convolution with layers Let's start with the ominous 32: ...
Broele's user avatar
  • 1,352
1 vote

Conv1d() input and output dimensions?

The number of input channels to a convolutional layer is given by the output of its previous layer. In this case, it is given by the output of the first 1D convolution (because ReLU is an element-wise ...
noe's user avatar
  • 25.7k
1 vote
Accepted

Questions about receptive field in the context of a practical CNN

Your drawing looks correct (assuming a stride of 1 and no dilation). We can talk about the receptive field for any layer in the CNN - so the receptive field of your 1st conv2d is 5 x 5, and of your ...
Lynn's user avatar
  • 1,274

Only top scored, non community-wiki answers of a minimum length are eligible