Reading material

This overview paper considers three areas: Gradient-based Learning (EM I am willing to make a presentation of this section: derivatives are not defined by a univeral property, such as the best linear approximation; it is unclear the correspondence between forward and reverse differentiation, the use of a type of linear maps seems a substantial improvement; lenses combine forward and backward propagation), Probability and Statistics (EM the main goal is to provide a high-level understanding of randomness, abstracting from the technicalities of measure theory), Invariant and Equivariant Learning (EM I did not read this section)