Current Situation

​Basically, we'll use MNIST as our test bench now.

​The preliminary goal we want to achieve is evolving an neural network that can rival the model in my NN_example repository, i.e. approximately 99.6 test accuracy in MNIST data set.

In the future, we will also adapt it into more data set, namely CIFAR-10, CIFAR-100 and ImageNet.

References

• Large-Scale Evolution of Image Classifiers
• Regularized Evolution for Image Classifier Architecture Search
• Learning Transferable Architectures for Scalable Image Recognition