Skip to content

intsystems/2026-Project-198

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Title

License GitHub Contributors GitHub Issues GitHub Pull Requests

Author Denis Gudkov
Consultant Daniil Dorin
Advisor Andrii Hrabovyi

Assets

Abstract

Modern vision models excel at recognition but fail to grasp geometric relationships like symmetry composition. This work investigates whether neural networks can internalize the algebraic structure of the dihedral group D₄ (square symmetries) rather than memorizing visual patterns. Using a Siamese encoder with autoregressive Transformer decoder, we train a model to predict whether two images are related by a D₄ transformation and identify the specific element. We demonstrate that the model learns true group properties: invariance to operation sequences (horizontal_flip → vertical_flip ≡ rotate_180), consistency with composition (g₂·g₁), canonical element representation, and correct rejection of unrelated pairs. Analysis of attention maps and embeddings reveals internal encoding of the D₄ multiplication table. Unlike VLMs that fail at such tasks, our architecture captures symbolic geometric structures.

Citation

If you find our work helpful, please cite us.

@article{citekey,
    title={Title},
    author={Name Surname, Name Surname (consultant), Name Surname (advisor)},
    year={2025}
}

Licence

Our project is MIT licensed. See LICENSE for details.

About

No description or website provided.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors