New answers tagged deep-learning
2
votes
When using class weights is bad?
Applying class weights (or resampling) to reach a more-balanced training dataset is usually done specifically to get more predictions of the rarer classes. That will very often cause a decrease in ...
- 12.2k
0
votes
Layer normalization details in GPT-2
Layer normalization (LN) and its implementation is one of the more confusing aspects of transformers.
First, the goal of normalization is to stabilize the gradient descent during training. ...
0
votes
How to binning/tokenizing amplitude of stationary timeseries?
This is straightforward of min-max scaling (normalization).
Except, the data is represented with unsigned integer after normalization. E.g. uint8 dtype can hold 2^8 or 256 vocabulary size.
2
votes
Why softmax training is more stable
The short answer is: Yes, it is easier to train a model using SoftMax for multiclass classification compared to sigmoid, and SoftMax generally yields better results (lower loss and higher accuracy).
...
- 121
0
votes
Daily Balance Prediction Using LSTM & ARIMA
This question pertains to market forecasts, but the answer may be useful in your case as well.
- 1
5
votes
Accepted
Which activation function for multi-class classification gives true probability (softmax vs sigmoid)
They don't represent true probability because you'd still have to calibrate your model.
Let's imagine you're trying to classify cats and dogs in a given set of images (binary classification problem, ...
0
votes
Decreasing reward when using DDPG
I am having the same problem when my actor loss becomes smaller but my reward reduces. Are you still following up this topic and have you figured it out the reasons? If yes, could you share with me?
...
0
votes
Why is my Keras model not learning image segmentation?
As per the comment by Pedro Henrique Monforte, since the OP has had plenty of time to do it themselves, I am thus turning the "answer" edited into the OP into an actual answer:
as is turns ...
- 1,988
3
votes
Accepted
Confidence levels and error rates in binary classification models
Neural networks are legendary for giving overconfident probability predictions that do not align with the reality of event occurrence. I have even heard people go so far as to say that the outputs are ...
- 4,244
Top 50 recent answers are included