Are there any theoretically-guided practices left in machine learning nowadays?

<!-- SC_OFF --><div class="md"><p>This wasn't always the case, but now almost any research topic in machine learning that you can imagine is now being done MUCH BETTER in industry

<!-- SC_OFF --><div class="md"><p>I’ve been reading about Geometric Deep Learning lately (the whole grids, graphs, groups, manifolds idea), and something clicked that i wanted to

<!-- SC_OFF --><div class="md"><p>Link: <a href="https://youtu.be/XExyqAYDnvw?si=BVyX-oEetFslAvZq&amp;t=8285">Judea Pearl, 2011 ACM Turing Award Recipient</a> (2:18:05)</p>

<!-- SC_OFF --><div class="md"><p>AI Engineering from Scratch is an MIT-licensed curriculum: 523 lessons across 20 phases, from linear algebra and backprop to transformers, LLMs,

<!-- SC_OFF --><div class="md"><p>The BDH (Dragon Hatchling) paper (arXiv:2509.26507) describes a Hebbian synaptic plasticity mechanism where model weights update during

<!-- SC_OFF --><div class="md"><p>Non-contrastive SSL methods like BYOL/JEPA/data2vec seem promising, but I have no idea what is being learned, or how well; it’s models all the

<!-- SC_OFF --><div class="md"><p>Hey everyone,</p> <p>I'm trying to get a better sense of where classic ML fits in the current landscape, dominated by LLMs and deep learning. Are

<!-- SC_OFF --><div class="md"><p>As someone who studied stats undergrad and industrial engineering operations research grad, and who thinks about the practical business of ML