Newbie training errors
During a recent personal project I was working with Deep Learning models, and in particular trying to fine tune a CNN (AlexNet) in Caffe. At first I was getting some pretty strange, and frustrating, learning curves.
It turned out that my error was that I was that I was running a script to augment my images prior to training, but forgetting to randomize the resulting dataset.
Luckily, once I corrected this, everything started nicely converging.
