That person approaches the issue as his personal problem. However, there are likely many other students around him with the same problem. It is a problem of the whole field recently.
One solution would be simply to a arrange a local seminar,
and understand a couple of papers in full detail. It would help to invite a couple of mathematically aware students, from mathematics, physics, or the part of cs
faculty where they prove stuff. They should be able to explain and answer questions immediately, which is way more effective than reading whole books or taking courses.
Those can be read for details later.
If the papers for the seminar are deep learning papers, part of the outcome of the seminar is likely to be an understanding that the authors of these papers do not necessarily understand the mathematics themselves.
One solution would be simply to a arrange a local seminar, and understand a couple of papers in full detail. It would help to invite a couple of mathematically aware students, from mathematics, physics, or the part of cs faculty where they prove stuff. They should be able to explain and answer questions immediately, which is way more effective than reading whole books or taking courses. Those can be read for details later.
If the papers for the seminar are deep learning papers, part of the outcome of the seminar is likely to be an understanding that the authors of these papers do not necessarily understand the mathematics themselves.