In this talk, I'd like to discuss the intertwining importance and connections of the three principles of data science in the title. In particular, we demonstrate the power of transfer learning from ImageNet data to neuron measurements collected by the Gallant Lab.
We employ the predictability and stability principles and use deep nets (CNNs) to understand pattern selectivities of neurons in the difficult primate visual cortex V4. We achieve state-of-the-art prediction performance, and obtain interpretations of diverse V4 neurons through stable "deep tune" visualizations over multiple predictive models.