New answers tagged deep-learning
0
votes
Is residual mapping always beneficial?
First of all, you can always construct an example to break assumptions.
"Say I take an MLP and generate data according to that MLP, then that MLP is the optimal solution, and anything you add on ...
- 2,473
0
votes
How to train an LSTM with varying length input?
This question might be deep learning framework specific, so my answer is intended for Keras.
In keras, an LSTM cell accept 3D tensor, which is ...
0
votes
Deriving ELBO for Diffusion Models
Indeed the notations here are a little sloppy and confusing since per definition of KL it's obvious that the two log terms within the 2nd and 3rd term of equation (44) should be rewritten as the two ...
- 3,029
0
votes
If I freeze pre-trained model weights and than train a classifier on top of its embeddings does that called fine-tunning?
I'd like to add another angle to @kostya's excellent answer.
I see the lack of use of additional input as a key distinction for not calling this adaptation fine-tuning. Moreover, as in today's ...
2
votes
Is it better to train neural network with ideal data before resuming with non-ideal data?
If you have the full dataset (comprising both "ideal"/cleaned and "noisy"/real examples), then no, it is not necessarily better to pretrain it with clean data just to initialise ...
- 121
0
votes
When to use Tanh?
Tanh:
can convert an input value(x) to the output value between -1 and 1. *0 and 1 are exclusive.
's formula is y = (ex - e-<...
1
vote
Accepted
What is the channel dimension other than color representation in Conv2D? Shall I use Conv3D instead?
You can do whatever you want, but it won't necessarily achieve a good result. CNNs have the inductive bias that features are spatially correlated. If you simply stack different views and immediately ...
- 85
0
votes
Deep Learning training strategy: Avoid shuffling individual training images, instead shuffle batches?
In Pytorch - DataLoader - If you set Shuffle to false , it will select the samples as they are stored - Sequential Access
Problems with Shuffle set to false : Gneralization, Overfitting.
If the model ...
- 11
Top 50 recent answers are included