Deep Learning - Exercise 09

Learning goal: Building an important CNN topology with TensorFlow

1. Prepare training and test data

Use ImageNet to download training and test images of two categories. Prepare at least 100 training and 100 test images for each category.

2. Write an AlexNet CNN in TensorFlow

In the lecture we introduced the AlexNet CNN architecture which was published in 2012 and is one reason for the large interest in the Deep Learning approach in machine learning due to its remarkable performance in the ILSVRC 2012 competition. See the ILSVRC 2012 results (AlexNet = Team "Supervision").

Rebuild the AlexNet CNN model in TensorFlow!

3. Experiment with your AlexNet implementation

Which is the best classification rate you can get with your implementation and the original parameter settings as described in the paper for a two class classification problem?