Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CASSANDRA: Programmatic and Probabilistic Learning and Inference for Stochastic World Modeling

Panagiotis Lymperopoulos, Abhiramon Rajasekharan, Ian Berlot-Attwell, Stéphane Aroca-Ouellette, Kaheer Suleman

TL;DR

This work proposes CASSANDRA, a neurosymbolic world modeling approach that leverages an LLM as a knowledge prior to construct lightweight transition models for planning that demonstrate significant improvements in transition prediction and planning over baselines.

Abstract

Building world models is essential for planning in real-world domains such as businesses. Since such domains have rich semantics, we can leverage world knowledge to effectively model complex action effects and causal relationships from limited data. In this work, we propose CASSANDRA, a neurosymbolic world modeling approach that leverages an LLM as a knowledge prior to construct lightweight transition models for planning. CASSANDRA integrates two components: (1) LLM-synthesized code to model deterministic features, and (2) LLM-guided structure learning of a probabilistic graphical model to capture causal relationships among stochastic variables. We evaluate CASSANDRA in (i) a small-scale coffee-shop simulator and (ii) a complex theme park business simulator, where we demonstrate significant improvements in transition prediction and planning over baselines.

CASSANDRA: Programmatic and Probabilistic Learning and Inference for Stochastic World Modeling

TL;DR

This work proposes CASSANDRA, a neurosymbolic world modeling approach that leverages an LLM as a knowledge prior to construct lightweight transition models for planning that demonstrate significant improvements in transition prediction and planning over baselines.

Abstract

Building world models is essential for planning in real-world domains such as businesses. Since such domains have rich semantics, we can leverage world knowledge to effectively model complex action effects and causal relationships from limited data. In this work, we propose CASSANDRA, a neurosymbolic world modeling approach that leverages an LLM as a knowledge prior to construct lightweight transition models for planning. CASSANDRA integrates two components: (1) LLM-synthesized code to model deterministic features, and (2) LLM-guided structure learning of a probabilistic graphical model to capture causal relationships among stochastic variables. We evaluate CASSANDRA in (i) a small-scale coffee-shop simulator and (ii) a complex theme park business simulator, where we demonstrate significant improvements in transition prediction and planning over baselines.
Paper Structure (43 sections, 6 equations, 2 figures, 5 tables, 3 algorithms)

This paper contains 43 sections, 6 equations, 2 figures, 5 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Framework Overview
  5. Modeling Deterministic Dynamics using LLM-Generated Code
  6. Code Components:
  7. Code Optimization:
  8. Modeling Stochastic Dynamics with LLM priors
  9. DAG structure search
  10. Sampling semantically coherent DAGs.
  11. Searching for DAGs with data and LLM prior.
  12. Fitting to data
  13. Experiments
  14. Experiment 1: CoffeeShopSimulator
  15. Results
  16. ...and 28 more sections

Figures (2)

  • Figure 1: Overview of the CASSANDRA architecture. Deterministic dynamics are modeled by LLM-generated code, which is optimized by an evolutionary algorithm using observed trajectories (top). Stochastic dynamics are modeled by a probabilistic graphical model, designed through simulated annealing structure search with an LLM prior and observed transitions (bottom).
  • Figure 2: Initial and final DAGs for CoffeeShopSim, modeling dependencies of state variables.