Active Inference Meeting Energy-Efficient Control
of Parallel and Identical Machines

Yavar Taheri Yeganeh · Mohsen Jafari · Andrea Matta

---

This project explores deep active inference, an emerging field that combines deep learning with the active inference decision-making framework. Leveraging a deep active inference agent, we focus on controlling parallel and identical machine workstations to enhance energy efficiency. We address challenges posed by the problem's stochastic nature and delayed policy response by introducing tailored enhancements.

• The project is based on a paper presented and accepted in the proceedings of the 10th International Conference on machine Learning, Optimization and Data science (LOD '24).

Key Contributions:

• Deep Active-Inference-Based Agent: With active inference integrating perception, learning, and action, the agent is tailored and enhanced for optimizing machine control in the stochastic manufacturing environment with delayed policy response.
• Hybrid Horizon and Multi-Step Transition Methods: These enhancements allow the agent to improve prediction and planning for longer horizons without computational burden.
• Performance: Experimental results show the effectiveness of enhancements compared to the existing agents as well as the potential of the proposed methodology. Notably, the agent achieved high rewards with simple one-step repeated multi-transition lookaheads.

Contents

This repository contains the implementations of the project, including source code for the agent as well as the manufacturing environment simulator used for experiments.

Requirements

You need a Python environment with the following libraries (and other supporting ones):

torch>=2.1.1
numpy>=1.24.3
simpy>=4.0.1

You can install them with pip by running:

pip install -r requirements.txt

Alternatively, you can use Conda with the YAML file provided in the repository:

conda env create -f tf-environment.yml

Running

Clone the source code by running:

git clone https://github.com/YavarYeganeh/AIF_Meeting_EEC.git
cd ./AIF_Meeting_EEC

Then, train the agent by running:

python train.py --batch <batch_size> --gamma <gamma_value> --steps <number_of_steps> --samples <number_of_samples> --calc_mean

Usage: train.py [-h]

• --batch : Select batch size, i.e., the number of environments used for training.
• --gamma : Select gamma value (between 0 and 1) for balancing between short and long horizons during planning.
• --steps : Number of actions the transition considers (in multi-step transitions).
• --samples : Number of samples to be used for Expected Free Energy (EFE) calculations.
• --calc_mean : Whether to consider the mean during EFE calculations (flag to include the mean).

Please note: $\lambda_s$ can be directly set in src/tfmodel.py, and by default in this source code, $\lambda_s = 1.5$.

After training is finished, the code records data related to the performance and statistics of the agent from the experiment in the path results/signature/stats_final.pkl.

Citation

Yeganeh, Y. T., Jafari, M., & Matta, A. (2024). Active Inference Meeting Energy-Efficient Control of Parallel and Identical Machines. arXiv preprint arXiv:2406.09322.

@article{yeganeh2024active,
  title={Active Inference Meeting Energy-Efficient Control of Parallel and Identical Machines},
  author={Yeganeh, Yavar Taheri and Jafari, Mohsen and Matta, Andrea},
  journal={arXiv preprint arXiv:2406.09322},
  year={2024}
}

Contact

For inquiries or collaboration, please reach out to yavar.taheri@polimi.it or yavaryeganeh@gmail.com.