Programming languages are our best descriptors of computation. They help us interpret, control, and construct machine behavior. Insofar as the brain computes, its best descriptor will also be a programming language. One which helps us interpret, control, and construct neural behavior. My goal is to find and build that language.
Currently, I see promise in strange and interesting programming language semantics. We can systematically derive compilers from these semantics. And these compilers yield programming languages whose programs compile to and from neural networks. I develop a first pass at the idea in this short proposal.
Joey Velez-Ginorio. Compositional desires as compositional programs. MSc Thesis, MIT (2021).
Joey Velez-Ginorio. Mapping programs to the brain. Proposal (2020).
Joey Velez-Ginorio, et al. Learning in System F. Preprint (2020).
Joey Velez-Ginorio. Learning in System F. MSc Thesis, University of Oxford (2019).
Joey Velez-Ginorio, et al. Interpreting actions by attributing compositional desires. 39th Annual Meeting of the Cognitive Science Society (2017).