Learning Human Habits with Rule-Guided Active Inference

Official repository for the paper "Learning Human Habits with Rule-Guided Active Inference" (ICLR 2026).

• Paper: OpenReview
• Authors: Zhiren Gong, Chao Yang, Wendi Ren, Shuang Li

Model framework (Figure 1)

Figure 1: Model framework. Wake–sleep rule learning and active inference.

---

Overview

Humans navigate daily life by combining deliberate planning in novel situations and fast, automatic responses in familiar ones. This work presents a framework for learning human habits with rule-guided active inference, where habits emerge as symbolic rules that serve as compact, interpretable shortcuts for action. We design a biologically inspired wake–sleep algorithm: in the wake phase, the agent harvests candidate rules from real experience; in the sleep phase, it performs generative replay to consolidate and prune rules under a unified free-energy objective. The framework supports both instant rule-based action in familiar scenarios and flexible planning via expected free energy in novel cases.

Main contributions:

• A biologically inspired extension of active inference for modeling human(-like) habits via rule-guided policies
• A novel wake–sleep algorithm that jointly learns generative models and symbolic rules
• Empirical evidence on NBA SportVU, car-following, medical diagnosis (DDXPlus), and Atari–Berzerk, with improved predictive performance and interpretability

---

Code

The code is currently under refinement. We will release the implementation as soon as it is ready. Please watch this repository or check back later for updates.

---

Citation

If you find this work useful, please cite:

@inproceedings{zhiren2026learning,
  title     = {Learning Human Habits with Rule-Guided Active Inference},
  author    = {GONG ZHIREN and Chao Yang and Wendi Ren and Shuang Li},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year      = {2026},
  url       = {https://openreview.net/forum?id=FZXwkBH6s7}
}

---

License

This project is for academic use. Please refer to the paper and the license file in the repository for details once the code is released.