license: mit
tags:
- world-model
- video-prediction
- hippocampus
- entorhinal-cortex
- representation-learning
HPC-MEC World Model
This repository hosts pretrained checkpoints for Structure Abstraction and Generalization in a Hippocampal-Entorhinal Inspired World Model.
The model is a hippocampal-entorhinal inspired world model that learns reusable transition structures from observation-only videos. It separates content-rich episodic representations from compact abstract dynamics, and uses velocity-like latent transitions for prediction and structural generalization across objects and scenes.
Paper
Structure Abstraction and Generalization in a Hippocampal-Entorhinal Inspired World Model
Tianqiu Zhang*, Muyang Lyu*, Xiao Liu, Si Wu
ICML 2026
- Paper: https://arxiv.org/abs/2605.15733
- Project page: https://hpc-mec-worldmodel.github.io/
Code
The training and evaluation code is available at:
https://github.com/senngadaisuki/hpc-mec-worldmodel
Usage
Please see the GitHub repository for installation, checkpoint loading, training, and evaluation instructions.
Citation
@inproceedings{zhang2026structure,
title = {Structure Abstraction and Generalization in a Hippocampal-Entorhinal Inspired World Model},
author = {Zhang, Tianqiu and Lyu, Muyang and Liu, Xiao and Wu, Si},
booktitle = {Forty-third International Conference on Machine Learning},
year = {2026},
url = {https://openreview.net/forum?id=AYXgo5FjYz}
}