A Practical Introduction to pymdp: A Two-Part Guide to Active Inference in Python

Approved: Yes
Title: A Practical Introduction to pymdp: A Two-Part Guide to Active Inference in Python
Category: Video
Skill Level:
Beginner (304)

Description:
Here are descriptions for a two-part video series from the Active Inference
Institute.  These videos introduce pymdp, a powerful Python package for
researchers and developers working with active inference and the Free Energy
Principle. 🧠 The first video provides a high-level overview of the
package, while the second dives into a practical coding demonstration.
Together, they offer a comprehensive guide from theory to application.

Active Inference ModelStream 007.1 ~ Conor Heins & Daphne Demekas ~ pymdp

This video provides an overview of pymdp, a Python package designed for
active inference. Active inference is a framework for modeling
decision-making and planning based on the Free Energy Principle, which posits
that intelligent agents act to minimize surprise or uncertainty about their
world. The pymdp package allows researchers to simulate how agents achieve
this through both perception and action. It was created as a more accessible,
user-friendly alternative to traditional MATLAB tools. The package is
modular, integrates with other Python libraries like PyTorch, and has future
plans to incorporate JAX for deep learning applications, making it a powerful
tool for neuroscience and AI researchers studying the Free Energy Principle.

Active Inference ModelStream 007.2 ~ pymdp

This video, the second part of a series, demonstrates how to implement an
active inference agent using the pymdp library. The presentation focuses on
building a generative model for a contextual multi-armed bandit task. Key
concepts explained include the four main components of a POMDP—the A, B, C,
and D matrices—which represent the agent's beliefs. The video also
introduces the concept of factorized state spaces to manage computational
complexity and details how an agent can learn by updating its model
parameters based on experience. The ultimate goal is to show how these
agents, following the Free Energy Principle, can engage in both optimal and
information-seeking behavior.



Link to Resource:
- Active Inference ModelStream 007.1 ~ Conor Heins & Daphne Demekas ~ pymdp (https://www.youtube.com/live/skf3sOM-7WI?si=ZqpSYMs8RXEUf77y)
- Active Inference ModelStream 007.2 ~ pymdp (https://www.youtube.com/live/uX8iSoDR83g?si=6BfOPlk1GSHpurIQ)

Tags:
ai (271), machine-learning (272), artificial-intelligence (884), cognitive-science (878), computer-science (875), python (69)

Domain:
ACCESS CSSN (780)

Would you like to associate this resource with an Affinity Group?: {Empty}