SOFT-DISKS FLUID LEARNING
AUTHOR: LUCA ZAMMATARO
DESCRIPTION: This work is based on the Equivalence between Molecular Dynamics and Neural Network. It provides learning proofs in a Lennard-Jones (LJ) fluid, presented as a network of particles having non-bonded interactions. I describe the fluid's learning as the property of an order that emerges as an adaptation in establishing equilibrium with energy and thermal conservation. The experimental section demonstrates the fluid can be trained with logic-gates patterns. The work goes beyond Molecular Computing's application, explaining how this model uses its intrinsic minimizing properties in learning and predicting outputs. Finally, it gives hints for a theory on real chemistry's computational universality.
LICENSE: BSD 3-Clause License
Usage: python LJSDF.py <training/testing file>
I.e.: python LJSDF.py AND.txt
Note: To "training" the algorithm with a logic-gate, use as argument one of the text files provided (AND.txt, NAND.txt OR.txt, NOR.txt XOR.txt, and XNOR.txt)
To "testing" a logic-gate, copy the 3 weights files (weights.r, weights.rv, and weights.ra) from the logic-gate directory you want to test, into the same position where you're launching the python script. Use as argument one of the text files into the LOGIC_PATTERNS directories.
The code is described in:
Molecular Learning of a Soft-Disks Fluid Luca Zammataro bioRxiv 2021.07.24.453642; doi: https://doi.org/10.1101/2021.07.24.453642