Why Do Conventional World Models Fail to Learn Cellular Automata?
cs.AI
Submitted: 2026-09-30
Updated: 2026-09-30
Code: https://github.com/guoshaoyang-pku/momentum-induction
Terminology
Sources
- Physics of Language Models: Part 4.1, Architecture Design and the Magic of Canon Layers
- Zoology: Measuring and Improving Recall in Efficient Language Models
- VideoPhy: Evaluating Physical Commonsense for Video Generation
- Relational inductive biases, deep learning, and graph networks
- Physion: Evaluating Physical Prediction from Vision in Humans and Machines
- LifeGPT: Topology-Agnostic Generative Pretrained Transformer Model for Cellular Automata
- AutomataGPT: Forecasting and Ruleset Inference for Two-Dimensional Cellular Automata
- Genie: Generative Interactive Environments
- Learning Elementary Cellular Automata with Transformers
- MaskGIT: Masked Generative Image Transformer
- Neural Ordinary Differential Equations
- Lagrangian Neural Networks
- The Evolution of Statistical Induction Heads: In-Context Learning Markov Chains
- What One Cannot, Two Can: Two-Layer Transformers Provably Represent Induction Heads on Any-Order Markov Chains
- Force Prompting: Video Generation Models Can Learn and Generalize Physics-based Control Signals
- Hamiltonian Neural Networks
- World Models
- Mastering Diverse Domains through World Models
- Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion
- Rotary Position Embedding for Vision Transformer
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems