Projective Simulation

Agency • Learning • Physics

Projective Simulation (PS) is a framework for studying agency from a physical perspective. PS agents learn from interaction with their environment and base their decisions on episodic memory. Internally, their deliberation corresponds to random walks on a clip network (graph), which provides an interpretable route to physical realizations and analysis.

This library provides all the necessary tools to develop, deploy, and train PS agents in a variety of environments. Its content ranges from entry-level tutorials introducing the core principles of PS to cutting-edge research on the latest developments and applications. It is actively developed by the QIC group at UIBK.

Installation

Projective simulation is available for python>=3.10 via

pip install projective_simulation

This will also install all required dependencies.

Getting Started

To begin exploring Projective Simulation, we recommend to start with the tutorials, which cover everything from the basics to more advanced applications.

If you wish to develop or deploy PS agents, please refer to the documetation.

Contributing

If you want to contribute to this library, please take a look at the contributing tutorial.

Citing

If you find this package useful in your projects, you can either cite the specific PS paper most relevant to your work (see a list here) or use the following to cite the library as a whole:

!soon!