Reprogram only the top layer in Keras AlexNet

Question

I'm still pretty new to Keras / Theano, so please bear with me. But after working with search and read documentation, I did not find anything (except writing my own optimizer)

I downloaded convnets from https://github.com/heuritech/convnets-keras , and here I wanted to use and retrain AlexNet. I just want to reinstall the upper layers (dense 1,2,3) for 8 classes instead of 1000, so it will save the “concepts” of images in conv layers, but it will generate output for my data (8 classes).

Is it possible to establish training only for the upper layers, for example. freeze lower layers?

thank