Neural Generative Coding (NGC) Experiments

Overview

This research project aims to replicate and evaluate the Neural Generative Coding (NGC) framework proposed by Ororbia and Kifer, The Neural Coding Framework for Learning Generative Models. The study provides a biologically inspired alternative to traditional backpropagation for training generative models.

Background

The NGC framework is rooted in predictive processing theory, offering a novel approach to machine learning that draws inspiration from biological neural systems. Unlike conventional training methods, NGC presents a unique perspective on how neural networks can learn and generate representations.

Research Objectives

• Replicating the NGC models described in the original study
• Evaluating model performance on benchmark datasets
• Comparing our results with the original study's findings
• Analyzing model performance across multiple evaluation metrics

Key Metrics of Evaluation

1. Reconstruction Quality: Measures how well the model reconstructs input data
2. Likelihood Estimation: Evaluates the model's ability to generate realistic samples
3. Classification Capability: Assesses whether learned representations are useful for downstream classification tasks

Installation

To set up the environment and run experiments, follow these steps:

1. Clone the Repository

git clone https://github.com/iCog-Labs-Dev/NAC-Experiments.git
cd NAC-Experiments/Model_Comparison/Stable_Version

2. Build the Docker Image

docker build -t nac-experiments .

3. Run the Container

docker run --rm -it nac-experiments

Citation

If you use this project in your research, please cite:

Ororbia, A. & Kifer, D. (2022). The neural coding framework for learning generative models. Nature Communications, 13(1), 2064. DOI: 10.1038/s41467-022-29632-7

@article{Ororbia2022,
  author={Ororbia, Alexander and Kifer, Daniel},
  title={The neural coding framework for learning generative models},
  journal={Nature Communications},
  year={2022},
  month={Apr},
  day={19},
  volume={13},
  number={1},
  pages={2064},
  issn={2041-1723},
  doi={10.1038/s41467-022-29632-7},
  url={https://doi.org/10.1038/s41467-022-29632-7}
}

License

ngc-learn is distributed under the BSD 3-Clause License.