Towards Computationally Efficient and Explainable Active Inference

This thesis explores how active inference, a powerful approach that emerged in neuroscience, can be applied to study decision-making in intelligent systems. Simulating dynamic environments, the results demonstrate that active inference outperforms traditional methods and demonstrates adaptability. We introduce several innovative techniques to make active inference algorithms efficient and explainable, enabling their use in diverse scenarios. These advancements open up exciting possibilities for developing intelligent, adaptable, and explainable systems, with potential applications in robotics, safe artificial intelligence, and bridging the gap between AI and biology.