Few years back I wrote a post about a side project about applying deep learning to visual puzzles solving (k10v.github.io).
Happy to learn now that it was cited by a recent paper, I hope that it shows that non-academic research is taken more seriously now.
https://arxiv.org/abs/2010.00763 [24] S. Kharagorgiev, “Solving bongard problems with deep learning,” k10v.github.io, 2020