Is There a Connection Between Clustering and Deep Learning?

One of the most disappointing results for me regarding deep learning was a paper which identified that deep new anetworks cannot learn shapes. Onew networks are usually more based on identifying textures. So it's like a method which can actually do this, identify shapes. And if we have this connection from camenes and delerning, and spectra clustering is also closely related to camins, so then what my hope is is that we can somehow work on the architecture of our model such that you get something which has a spectral clustering component,. Then it will hence learn to classify by shapes.