Neuro-Symbolic AI

Neuro-symbolic AI systems span a spectrum ranging from those that simply combine separate "neuro" and "symbolic" systems while leaving each intact, which are the easiest to understand (and there are advantages to this kind of combination) to those that more fuse concepts from both sides in deeper ways, which can open up powerful new opportunities, but also many questions: What is the essence of what it means to be a 'neural' approach? What is the essnce of what it means to be a 'symbolic' approach? Can one formalism actually be both at the same time? What tend to be the strengths and limitations of each approach, and why? What are the systematic theoretical and empirical ways to characterize these strengths and limitatons? Which properties tend to trade off with each other, either formally or empirically? What are the deeper principles underlying each of these that can idneitfy and utilize to create entirely new kinds of models that achieve more of our desiderata, or at least better tradeoff points?

Another barrier to entry is the accelerating pace of new approaches, voices, and themes within neuro-symbolic AI. We approach these two barriers by maintaining a large network of experts at the top of their fields throughout all of the sub-disciplines involved so that we can present a curated selection that is refreshed each year, balancing ideological and technical coverage with the time cost of learning entirely new areas for practicing researchers.