BK
Dropout for regularisation
Size: a a a
2019 July 23
LS
Okay will try this too. You know any research papers supporting the use of NN for a small dataset
LS
As I need to defend using NN if it works at all
BK
Okay will try this too. You know any research papers supporting the use of NN for a small dataset
Can send mine lol
LS
Is it published?
BK
Used convolutions with 400 samples lol kek
LS
Indexed in Scopus ?
DP
Used convolutions with 400 samples lol kek
thats unfair, convolutions can have really small kernels, so amount of parameters can be really low
BK
Is it published?
Yup, in scopus
LS
Okay I will find your paper thanks
BK
thats unfair, convolutions can have really small kernels, so amount of parameters can be really low
True
BK
Okay I will find your paper thanks
BK
Here is the preprint
BK
Can't locate the proper one from the phone
BK
Just google scholar it to cite
LS
Thanks reading this now. You guys ar every helpful
DP
3 layers of 150 neurons fully connected layers = 67k parameters(150^2*3). need to delete >99% of the weak weights (lowest absolute value) to avoid overfitting
LS
BK
*very
Also you can add stuff like differential plasticity to improve your learning
LS
Is this part of NN?