DeepDFA: Injecting Temporal Logic in Deep Learning for Sequential Subsymbolic Applications
Elena UmiliSapienza University of RomeItaly [email protected] Francesco ArgenzianoSapienza University of RomeItaly [email protected] Roberto CapobiancoSony AISwitzerland [email protected] · 2026
Read DeepDFA: Injecting Temporal Logic in Deep Learning for Sequential Subsymbolic Applications by Elena UmiliSapienza University of RomeItaly [email protected] Francesco ArgenzianoSapienza University of RomeItaly [email protected] Roberto CapobiancoSony AISwitzerland [email protected] (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03486.txt.
Description
Teaching a neural network a rule that unfolds over time, like a sequence of required steps, is notoriously awkward. Elena Umili and 2 co-authors present DeepDFA in a 2026 paper, a neurosymbolic framework that injects temporal logic into deep learning. The rules are written as deterministic finite automata or Moore machines, the classic formal models of step-by-step behavior.
DeepDFA turns those automata into continuous, differentiable layers, so gradient training can respect the symbolic rules while learning from raw observations. The authors test it on classifying static image sequences and on policy learning in environments where rewards depend on history. Across experiments it beats recurrent networks, transformers, and rival neurosymbolic systems on temporal knowledge integration.
Catalog record
| CREATOR | Elena UmiliSapienza University of RomeItaly [email protected] Francesco ArgenzianoSapienza University of RomeItaly [email protected] Roberto CapobiancoSony AISwitzerland [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-03486