groupoid-rl
Introduction
Properties of the underlying group has been used to train RL to partially solve a rubik's cube [1]. We believe that we can generalize this result from groups to groupoids.
Groupoids
Groupoids have an algebraic and a category theory definition. We focus on the algebraic definition for this article.
Training
Run
python train.py
Both the state of the art resnet model from the DeepCubeA paper and our groupoid model will be trained.
graph LR
A[python train.py] -- Trains both --> B((Groupoid Equivariant ))
A --> C(Resnet)
B --> D{Results and figures }
C --> D
Slide-15
[1] Fourier Bases for Solving Permutation Puzzles
[2] SNPY
[3] Groupoid Article
[4] DeepCubeA
[5] [Groupoids][https://groupoids.org.uk/pdffiles/groupoidsurvey.pdf]
[6] slide-15
[7] Groebner Basis
[8] Groupoid N-lab
[9] [Groupoid Basis Hilbert Space] (https://arxiv.org/pdf/1907.09010.pdf)