The Modern Mathematics of Deep Learning

It seems like this can leave the reader with the wrong impression. Calculus really is &quot;the mathematics of Newtonian physics&quot;. This is just &quot;some mathematics that might help a bit in your intuitions of deep learning&quot;.<p>IE, Deep learning is fundamentally just about getting the mathematically simple but complex and multi-layerd &quot;neural networks&quot; to do stuff. Training them, testing them and deploying them. There are many intuitions about these things but there&#x27;s no complete theory - some intuitions involve mathematical analogies and simplifications while other involve &quot;folk knowledge&quot; or large scale experiments. And that&#x27;s not saying folks giving math about deep learning aren&#x27;t proving real things. It&#x27;s just they characterizing the whole or even a substantial part of such systems.<p>It&#x27;s not surprising that a complex like a many-layered Relu network can&#x27;t fully characterized or solved mathematically. You&#x27;d expect that of any arbitrarily complex algorithmic construct. Differential equations of many variables and arbitrary functions also can&#x27;t have their solutions fully characterized.

Tangent, but has anyone taken Fast.ai or similar courses and transitioned into the Deep Learning&#x2F;ML field without a MS&#x2F;PhD? To be honest, I don&#x27;t even know what &#x27;doing ML&#x2F;DL&#x27; looks like in practice, but I&#x27;m just curious if a lot of folks get in to the field without graduate degrees.

What are the prerequisites?

After skimming through the paper it&#x27;s clear that the title should be read as &quot;The Modern (Mathematics of Deep Learning)&quot; and not my original parse which was &quot;The (Modern Mathematics) of Deep Learning.&quot; Very different things.

Wake me up when &#x27;deep learning&#x27; has independently created a language to communicate within a group of peers while under environmental pressure.<p>(and that language is co-expressive with human languages)

I believe that the curse of dimensionility doesn&#x27;t apply here as we are optimizing the &quot;universal apppriximator&quot; of the &quot;surface&quot; of the possible real world function.