Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Agentic Framework for Neuro-Symbolic Programming

Aliakbar Nafar, Chetan Chigurupati, Danial Kamali, Hamid Karimian, Parisa Kordjamshidi

TL;DR

ADS presents an agentic, multi-agent workflow to synthesize complete neuro-symbolic programs from natural language task descriptions, bypassing library-specific syntax and enabling optional human refinement. By integrating RAG grounding, modular knowledge and model declarations, and a lightweight sensor/LLMModel architecture, ADS reduces NeSy development time while maintaining semantic correctness through iterative self-refinement and a human-in-the-loop. Experimental results across 12 tasks show strong graph correctness for several LLMs (e.g., up to 97.22% C+R in knowledge declarations) and substantial practical gains in development time for both experts and non-users. The approach demonstrates a scalable path to rapid NeSy programming and suggests future extensions to other frameworks and automatic component extraction for open-source symbolic systems.

Abstract

Integrating symbolic constraints into deep learning models could make them more robust, interpretable, and data-efficient. Still, it remains a time-consuming and challenging task. Existing frameworks like DomiKnowS help this integration by providing a high-level declarative programming interface, but they still assume the user is proficient with the library's specific syntax. We propose AgenticDomiKnowS (ADS) to eliminate this dependency. ADS translates free-form task descriptions into a complete DomiKnowS program using an agentic workflow that creates and tests each DomiKnowS component separately. The workflow supports optional human-in-the-loop intervention, enabling users familiar with DomiKnowS to refine intermediate outputs. We show how ADS enables experienced DomiKnowS users and non-users to rapidly construct neuro-symbolic programs, reducing development time from hours to 10-15 minutes.

An Agentic Framework for Neuro-Symbolic Programming

TL;DR

ADS presents an agentic, multi-agent workflow to synthesize complete neuro-symbolic programs from natural language task descriptions, bypassing library-specific syntax and enabling optional human refinement. By integrating RAG grounding, modular knowledge and model declarations, and a lightweight sensor/LLMModel architecture, ADS reduces NeSy development time while maintaining semantic correctness through iterative self-refinement and a human-in-the-loop. Experimental results across 12 tasks show strong graph correctness for several LLMs (e.g., up to 97.22% C+R in knowledge declarations) and substantial practical gains in development time for both experts and non-users. The approach demonstrates a scalable path to rapid NeSy programming and suggests future extensions to other frameworks and automatic component extraction for open-source symbolic systems.

Abstract

Integrating symbolic constraints into deep learning models could make them more robust, interpretable, and data-efficient. Still, it remains a time-consuming and challenging task. Existing frameworks like DomiKnowS help this integration by providing a high-level declarative programming interface, but they still assume the user is proficient with the library's specific syntax. We propose AgenticDomiKnowS (ADS) to eliminate this dependency. ADS translates free-form task descriptions into a complete DomiKnowS program using an agentic workflow that creates and tests each DomiKnowS component separately. The workflow supports optional human-in-the-loop intervention, enabling users familiar with DomiKnowS to refine intermediate outputs. We show how ADS enables experienced DomiKnowS users and non-users to rapidly construct neuro-symbolic programs, reducing development time from hours to 10-15 minutes.
Paper Structure (34 sections, 3 figures, 5 tables)

This paper contains 34 sections, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background: DomiKnowS
  3. System Overview
  4. RAG
  5. Knowledge Declaration
  6. Model Declaration
  7. User Interface Implementation
  8. Backend
  9. Frontend
  10. Experiments
  11. Datasets
  12. Language Models
  13. Evaluation of ADS Workflow
  14. Human Evaluation
  15. Related Work
  16. ...and 19 more sections

Figures (3)

  • Figure 1: Knowledge and model declarations of a DomiKnowS program for the WIQA task. The left side defines the conceptual graph and logical constraints, while the right side shows the sensor code, which specifies how properties and predictive models are attached to the graph's concepts.
  • Figure 2: ADS system workflow. The left panel groups components of the Knowledge Declaration phase and the right panel the Model Declaration phase. Rectangles denote LLM agents, dashed rectangles denote human actions, ellipses denote input/output, and diamonds denote decision points.
  • Figure 3: The Graph Human Reviewer Interface. The dashboard presents the latest generated graph draft (left) alongside the verdicts from the Reviewer and Execution agents. The user can track the agent workflow history in the "Live Progress" panel (right) and either approve the code or provide natural language feedback for revision.