SOCIA-Nabla: Textual Gradient Meets Multi-Agent Orchestration for Automated Simulator Generation

2025-10-21

AI Summary

This work addresses the challenges of expert-dependent simulator construction, fragile prompt engineering, and poor reproducibility. Methodologically, it formulates simulator generation as an optimization problem over code instances, introducing a text-based gradient descent mechanism that unifies multi-agent collaborative orchestration and loss alignment. Leveraging a closed-loop pipeline—code synthesis, execution, evaluation, and repair—augmented by human-in-the-loop validation, the framework enables end-to-end trainable simulation code generation. Its key contribution lies in being the first to integrate large language model–driven agent systems with textual computational graphs, rendering executable code itself a differentiable object. Evaluated on three cyber-physical system tasks—user modeling, mask adoption, and individual mobility—the approach achieves state-of-the-art accuracy while supporting cross-domain and multi-granularity simulation generation.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Multiagent Systems: Agent-Based Simulation and Emergent Behavior

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Agentic search

Abstract

In this paper, we present SOCIA-Nabla, an end-to-end, agentic framework that treats simulator construction asinstance optimization over code within a textual computation graph. Specialized LLM-driven agents are embedded as graph nodes, and a workflow manager executes a loss-driven loop: code synthesis -> execution -> evaluation -> code repair. The optimizer performs Textual-Gradient Descent (TGD), while human-in-the-loop interaction is reserved for task-spec confirmation, minimizing expert effort and keeping the code itself as the trainable object. Across three CPS tasks, i.e., User Modeling, Mask Adoption, and Personal Mobility, SOCIA-Nabla attains state-of-the-art overall accuracy. By unifying multi-agent orchestration with a loss-aligned optimization view, SOCIA-Nabla converts brittle prompt pipelines into reproducible, constraint-aware simulator code generation that scales across domains and simulation granularities. This work is under review, and we will release the code soon.

Problem

• Automating simulator construction through multi-agent code optimization
• Converting brittle prompt pipelines into reproducible simulator generation
• Scaling constraint-aware code generation across domains and granularities

Innovation

• Multi-agent orchestration optimizes code via textual computation graph
• Textual-Gradient Descent enables loss-driven code synthesis and repair
• Human-in-the-loop interaction minimizes expert effort in simulator generation