Build an image classifier by a convolution neural network and perform training/testing in the AWS AMI instance.
The dataset CIFAR-10 is downloaded into the linux server from CIFAR-10 unpacked and read into the numpy.array
AWS AMI Ubuntu 16.04 with tensorflow, keras pre-installed. A jupyter notebook server is configured on the Linux server by the following tutorial from AWS .
The notebook is opened in the local linux machine and computation is done by the server.
Training performance with the simple ConvNN can reach ~90% accuracy but the testing performance reaches ~70%. The 20% gap denotes the training dataset has overfit issues.