To realize the goal of "understanding language by machines", I think it's crucial to understand the logics behind that language. Since traditional approaches to logics have been brittle and suffers from low coverage, a machine learning-based approach is needed. I view it not as a replacement but a complement to traditional approaches. We will need both to solve the puzzle someday.
This entry is an ongoing effort to survey current literature on machine learning-aided logics. Use the "Follow" button to receive notifications when it is updated.
- Atoms as matrix factorization
- Gradient-based logics
- Knowledge base completion: how to induce new knowledge