A simple template tasks that helps zo understand the benchmark setup.
We want to focus on easily available and popular datasets.
For the template we create random data
A simple MLP will try to adapt to the random data.
There is no actual performance measured here. The loss is logged.