Limitations of Deep Learning

 

A couple of years ago, Gary Marcus wrote a paper: Deep Learning: A Critical Appraisal. Today, I believe similar concerns still remain valid. Hybrid architectures that involve (abstract) knowledge bases appear to be the elusive solution to the algorithmic generalization issue. Currently, one of my projects is a case study i.e. proof of concept for a hybrid deep learning / knowledge-based modeling environment. One of the challenges is the selection of the training set. I’m leaning towards a legal focus since the regulatory legislation aiming at AI systems is near. We are looking at a lot of fuzzy and nonmonotonic logic reasoning modules as an addition to ML.

https://syncedreview.com/2019/02/15/ai-hasnt-found-its-isaac-newton-gary-marcus-on-deep-learning-defects-frenemy-yann-lecun/