 So what did we learn today? We learned about multi-layer perceptrons and the importance of non-linearities. We learned how to construct them. We learned a little bit about their dynamics in the large width limit, and we learned how activation function matters in this context. And we also learned how to apply MLPs properly to real data sets. And I think this is a skill we will be learning using a lot for the remainder of the time. Now, I want to make sure that you really submit the results. We want you to use the air table so that we get all the data so that we can really learn how we can make this cost better. Thank you very much for submitting and thank you for being with us.