SeL-NAL

This repo contains the source code of the paper A Brain-inspired Sequence Learning Model based on Non-Axiomatic Logic.

Compilation

This project is python-C++ mixed. The C++ code is compiled by g++, with the standerd C++20; older C++ standards (e.g., c++11) and other compilers (e.g., clang) are not ensured to work.

To compile the C++ code, run the commands as the following

git submodule update --init --recursive
mkdir build
cd build
cmake ..
make

There would be two target folders, ./narsese and ./SequentialGroup, which are also two python modules.

Run

To test the capacity of the model, run the command

python -m test_capacity

To test the catastrophic forgetting phenomenon of the model, run the command

python -m test_catastrophic_forgetting

Run the jupyter-notebook tests.ipynb to get all related figures from tests.

Debug

Python-C++ mixed debugging is avaiable in VS-Code. To do so, the extension Python C++ Debugger is needed. Select the task "Python C++ Mixed" in .vscode/launch.json, and run and debug.

To enter a C++ file, a break point should be set up before debugging.

Performance

Capacity

length=5, n_patterns=5, n_types=26

length=14, n_patterns=20, n_types=26

length=14, n_patterns=20, n_types=1000

Continual Learning (without Catastrophic Forgetting)

length=10, n_patterns=10, n_types=26, n_repeats=3, n_episods=3

Citation

This work is still under peer review.

Cite as

@article{xu2023selnal,
  title={A Brain-Inspired Sequence Learning Model based on a Logic},
  author={Xu, Bowen},
  journal={arXiv preprint arXiv:2308.12486},
  year={2023}
}

(bibtex)