ReLU and Softplus Neural Nets as Zero-Sum, Turn-Based, Stopping Games
By Yiannis Vlassopoulos
A Computable Measure of Suboptimality for Entropy-Regularised Variational Objectives
By Anna Korba
Appears in collection : Mikefest : A conference in honor of Michael Douglas’ 60th birthday
Here I will discuss the frontier of research for formal reasoning via deep neural networks. I will highlight the most recent progress in the area, especially automated theorem proving and automated formalization of natural language text. Also, I will discuss the role of language models, contrastive training, retrieval augmented modeling, and reinforcement learning toward the long-term goal of creating a system that can formalize and verify theories starting from a large natural language corpus.