Logic Tensor Networks (LTN)

Dependencies

The following is what we are using for development. Basically similar versions should run fine.

• python3.7
• tensorflow >=2.0 (for running the core, wrapper etc)
• numpy >= 1.17.2 (for examples and tests)
• matplotlib >= 3.1.1 (for examples)

Installing dependencies is easy. Just use pip install tensorflow numpy matplotlib or use a virtualenv.

Repository structure

• logictensornetworks.py -- core system for defining constants, variables, predicates, functions and formulas.
• logictensornetworks_wrapper.py -- a simple wrapper that allows to express constants, variables, predicates, functions and formulas using strings.
• logictensornetworks_library.py -- a collection of useful functions.
• examples_ltn -- examples using the core system
• examples_ltnw -- examples using the wrapper
• tests -- tests

Running tests

Tests are in tests and should be run from the project root. To run all available tests use python3.6 tests/_all.py.

Currently, tests are for the wrapper.

Running examples

There are various examples for LTN core examples_ltn and how to use the wrapper examples_ltnw.

Run examples from the project root, e.g. python3.6 examples_ltn/multilable_classifier_simple.py

Papers

• Logic Tensor Networks: Deep Learning and Logical Reasoning from Data and Knowledge, Luciano Serafini and Artur d'Avila Garcez, Arxiv.org
• Learning and Reasoning with Logic Tensor Networks, Luciano Serafini and Artur d'Avila Garces, Proc. AI*IA 2016
• Learning and reasoning in logic tensor networks: theory and application to semantic image interpretation, Luciano Serafini, Ivan Donadello, Artur d'Avila Garces, Proc. ACM SAC 2017
• Logic tensor networks for semantic image interpretation, Ivan Donadello, Luciano Serafini and Artur d'Avila Garces. Proc. IJCAI 2017

Tutorias

Checkout recent tutorials on Logic Tensor Networks (LTN)

• IJCNN 2018 tutorial
• IJCAI 2018 tutorial

Other resources

• What are “Logic Tensor Networks”?, Lucas Buchberger
• Dagstuhl Seminar “Human-Like Neural-Symbolic Computation, Lucas Buchberger
• Human-Like Neural-Symbolic Computing. 2017, Dagstuhl Seminar 17192

License

This project is licensed under the MIT License - see the LICENSE file for details

Acknowledgments

LTN has been developed thanks to active contributions and discussions with the following people:

• Alessandro Daniele (FBK)
• Artur d’Avila Garces (City)
• Francesco Giannini (UniSiena)
• Giuseppe Marra (UniSiena)
• Ivan Donadello (FBK)
• Lucas Brukberger (UniOsnabruck)
• Luciano Serafini (FBK)
• Marco Gori (UniSiena)
• Michael Spranger (Sony CSL)
• Michelangelo Diligenti (UniSiena)