MNIST handwritten digits recognition using a Convolutional Neural Network in DeepLearning4J in Scala (taken from its Java original)
Some hyper-parameter tuning on the original version. This subdirectory has a version with an extra layer, a batch normalization layer in the CNN for MNIST. The initializers and data samplers (seeds, etc) are random.
To run, use, with SBT (Simple Build Tool), do a:
$ sbt run
...
==========================Scores========================================
Accuracy: 0.9898
Precision: 0.9898
Recall: 0.9897
F1 Score: 0.9897