OCANNL, the `neural_nets_lib`. Mysteries of NNs training unveiled in OCaml

Using OCANNL, we will build a toy feed forward network, we will train it, visualize its outputs. We will take a peek at the actual computation generated at various levels of abstraction: tensor node assignments, optimized C-language-like programs, translations to actual C (for CPUs, or CUDA but it is still work-in-progress). Lastly, we will parallelize training across multiple devices.