Frog

Frog makes QEC.

Frog is a codebase for experiments with neural decoder for Toric code.

The implementation is organized around a modular decoding pipeline: a code, a noise channel, and a decoder. Each of these components is defined through an abstract interface, and concrete implementations follow these interfaces. It is worth preserving this structure when extending the codebase.

The modules experiment, modules, net_modules, loss, and trainer contain the code related to training learnable decoders.

How to use

You can explore the examples in the examples_notebook/ folder. These notebooks demonstrate the main routines and objects from the codebase, including code construction, symmetry actions, coset decoding, and threshold evaluation.

Example notebooks

The examples_notebook/ folder contains small Jupyter notebooks demonstrating the main components of Frog and the experiments used for equivariant neural decoding of quantum error-correcting codes.

• _draw_code.ipynb
  Visualizes the lattice/code structure and the objects used by the decoder. Useful as a first notebook for understanding the geometry of the problem.

• _symmetries_action.ipynb
  Demonstrates how code symmetries act on syndromes, errors, and decoder inputs. This is the main notebook for understanding the equivariance structure used in END.

• _test_translation.ipynb
  Checks translation actions and verifies that the implemented symmetry transformations behave consistently.

• _coset_decoder.ipynb
  Shows how the coset decoder is constructed and tested. Useful for comparing neural decoding against a structured/reference decoding procedure.

• threshold_evaluation.ipynb
  Runs threshold-style evaluation experiments for decoders across different physical error rates and code sizes.

Paper

This repository is related to the paper:

• The END: An Equivariant Neural Decoder for Quantum Error Correction

Many thanks to Roberto Bondesan and Max Welling.