Invariants for Neural Automata
2023
Invariants for Neural Automata
publication
Evidence: high
Author Information
Author(s): Uria-Albizuri Jone, Carmantini Giovanni Sirio, beim Graben Peter, Rodrigues Serafim
Hypothesis
What aspects of the dynamics observed under a Gödel representation are intrinsic to the dynamics and which are not?
Conclusion
The study provides a rigorous mathematical framework for investigating symmetries and invariants of neural automata under different encodings.
Supporting Evidence
- The study defines patterns of equality for neural automata.
- Only step functions defined over patterns of equality are invariant under symbolic recodings.
- Mean activation levels are not invariant under these recodings.
Takeaway
This study looks at how different ways of encoding information in neural networks can affect the results we get, and it finds some things that stay the same no matter how we encode them.
Methodology
The study develops a mathematical framework to analyze the invariance properties of neural automata under different Gödel encodings.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website