Mini batch loss in CNN is almost constant and jumps around 0.69
1 view (last 30 days)
Show older comments
I have designed a CNN to work with seizure detection from EEG. My conv filters sizes are: 1x64 and 18x1 respectively.
imageInputLayer([18 64 3]); convlayer1; reluLayer(); dropoutLayer(0.1); convlayer2; reluLayer(); dropoutLayer(0.1) fullyConnectedLayer(2); dropoutLayer(0.2); softmaxLayer(); classificationLayer()];
When I train my CNN, the mini-batch loss almost hovers around 0.69 and doesn't get reduced even after 30 epochs.I tried reducing the filter size and adding Maxpool layers as well, still, the results are same. the maximum accuracy I am able to reach is 62%.
Can anyone help me with this problem? Thanks for your time and help.
Regards Venkat
Answers (1)
shefali saxena
on 19 Jan 2019
Hello sir
I am facing the same problem when training the CNN architecture using ECG signal
I tried with many epochs, changing layers and also by changing parameters but my training loss (mini-batch loss) doesn't decrease. and remains constant around a value of 0.69 and accuracy not improving beyond 50%
how to increase accuracy
Please help!!!!
Thanks and regads.
0 Comments
See Also
Categories
Find more on Text Analytics Toolbox in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!