Research · AI · NLP
LTNN
Logical Transformer Neural Network
A network that combines Logical Neural Networks with transformers: logical masking in attention, bidirectional reasoning, and more explainability for NLP.
Problem
A standard transformer does not declare logical constraints. In ambiguity, KBQA, or coreference, it is hard to audit why attention ignores a rule the domain actually has.
Approach
LTNN (author: Mauricio Pacheco Lizama) inserts real-valued logic and constraints into attention, with bidirectional reasoning. The goal is a model that can respect — and show — logical structure, not only predict the next token.
Outcome
An open-source research project. We do not invent GLUE or KBQA scores; the documented value is the architecture and the repository.
Stack
Python · Transformers · Logical Neural Networks · NLP
Services