M

MachineLearningCompphys

Settings | Report Duplicate

1

I Use This!

Very Low Activity

Commits : Listings

Analyzed 2 days ago. based on code collected 2 days ago.

Commit Message	Contributor	Files Modified	Lines Added	Lines Removed	Code Location	Date
Oct 14, 2024 — Oct 14, 2025 Showing page 1 of 7 Search / Filter on:
Added blog page reference to README.md	Adrian Roman	More...				about 1 year ago
Fixed typo in function naming	Adrian Roman	More...				over 1 year ago
Replaced some typedefs with using	Adrian Roman	More...				over 1 year ago
Commented out the Gaussian noise parameter, since it doesn't seem to help for this particular case (dropout on input + batch normalization).	Adrian Roman	More...				over 1 year ago
Added a comment about dropout with batch normalization	Adrian Roman	More...				over 1 year ago
Experimental code (currently testing it) for adding Gaussian noise to input and the output of the hidden layers. Unlike dropout, this might work ok together with the batch normalization.	Adrian Roman	More...				over 1 year ago
Wrong assert preventing compiling in debug mode. Minor other changes	Adrian Roman	More...				almost 2 years ago
Updated README.md with the latest results	Adrian Roman	More...				almost 2 years ago
Better pretrained ensemble using AdamW solver instead of Adam and also a better value for beta2 parameter. Getting almost 99.55% accuracy on the test set.	Adrian Roman	More...				almost 2 years ago
Dropped the default beta2 value to 0.98, the previous default (0.999, as in the paper) looks too much for more complex situations.	Adrian Roman	More...				about 2 years ago
Still playing with hyperparameters	Adrian Roman	More...				about 2 years ago
Still playing with the params	Adrian Roman	More...				about 2 years ago
Just changed the order of some params initialization, nothing important	Adrian Roman	More...				about 2 years ago
Another comment corrected	Adrian Roman	More...				about 2 years ago
Corrected a typo in a comment for AdamW	Adrian Roman	More...				about 2 years ago
Fixed an issue with parameters decay for AdamW, still need to test	Adrian Roman	More...				about 2 years ago
Removed the regularization term for bias from AdamW	Adrian Roman	More...				about 2 years ago
Added AdamW in README.md	Adrian Roman	More...				about 2 years ago
Added AdamW https://github.com/aromanro/MachineLearning/issues/29 and used it as default everywhere instead of Adam. Not tested yet, though.	Adrian Roman	More...				about 2 years ago
Update README.md	Adrian Roman	More...				about 2 years ago
Minor changes	Adrian Roman	More...				over 2 years ago
Added function that allows saving an EMNIST image (or an output from a neural network) to a PPM file	Adrian Roman	More...				over 2 years ago
Moved out a loading data function from NeuralNetworkTestsMNIST.cpp	Adrian Roman	More...				over 2 years ago
Corrected parenthesis for Eigen link in README.md	Adrian Roman	More...				over 2 years ago
Added info in README.md about Eigen	Adrian Roman	More...				over 2 years ago
Moved ensemble testing in a separate function.	Adrian Roman	More...				over 2 years ago
Added info in the README.md about the ensemble performance.	Adrian Roman	More...				over 2 years ago
Added ensemble: 99.4575% on the test set while no 'pretrained' network from ensemble reaches 99.4% (highest is at 99.3725%, lowest is 99.2875%)	Adrian Roman	More...				over 2 years ago
Small changes	Adrian Roman	More...				over 2 years ago
Added some more things in README.md	Adrian Roman	More...				over 2 years ago

←
1
2
3
4
5
6
7
→