FUTURE_REVISIONS.txt

# Future Revisions to the Notes

The following revisions are recommended in advance of a future
semester. In some cases these are extensive or cross multiple chapters
and weekly assignments, so need to be made in a coordinated fashion
rather than during the semester. In other cases, these are changes
noticed after the notes had been released and with assignments
depending on them, so the changes have been deferred.

## Notation

* In neural network structures, $m^l$ is used as number of inputs for
layer $l$, and $n^l$ as the number of outputs. This clashes with the
use of $n$ as the number of training samples. Better might be $d$ for
the number of inputs (consistent with $d$ previously being the number
of features associated with a training sample), and perhaps $m$ for
the number of outputs.

* Also in the neural networks chapter, x for the data is now sometimes
x and sometimes X. The difference between a and A is similarly a bit
unclear.

## Specifics

* Should define what a "feature" is, in the intro chapter.

* The figure on p. 55 showing gradient descent should be revised to be
more consistent between contours and gradients.