Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DeepInflation: an AI agent for research and model discovery of inflation

Ze-Yu Peng, Hao-Shi Yuan, Qi Lai, Jun-Qian Jiang, Gen Ye, Jun Zhang, Yun-Song Piao

TL;DR

DeepInflation addresses the challenge of exploring the expansive inflationary model space by fusing LLM-based reasoning, symbolic regression, and a retrieval-augmented knowledge base grounded in the Encyclopaedia Inflationaris. The approach autonomously discovers simple single-field slow-roll potentials that satisfy observational constraints on $n_s$ and $r$ (e.g., ACT DR6) and provides theoretical context for a broad set of inflationary scenarios. It demonstrates capabilities for both model discovery and literature-informed model inquiries, enabling researchers and non-experts to navigate the inflationary landscape through natural language. The work outlines a path toward autonomous scientific discovery in cosmology, while noting limitations to single-field models and outlining future extensions to multi-field dynamics, PBH production, and fully autonomous closed-loop operation.

Abstract

We present \textbf{DeepInflation}, an AI agent designed for research and model discovery in inflationary cosmology. Built upon a multi-agent architecture, \textbf{DeepInflation} integrates Large Language Models (LLMs) with a symbolic regression (SR) engine and a retrieval-augmented generation (RAG) knowledge base. This framework enables the agent to automatically explore and verify the vast landscape of inflationary potentials while grounding its outputs in established theoretical literature. We demonstrate that \textbf{DeepInflation} can successfully discover simple and viable single-field slow-roll inflationary potentials consistent with the latest observations (here ACT DR6 results as example) or any given $n_s$ and $r$, and provide accurate theoretical context for obscure inflationary scenarios. \textbf{DeepInflation} serves as a prototype for a new generation of autonomous scientific discovery engines in cosmology, which enables researchers and non-experts alike to explore the inflationary landscape using natural language. This agent is available at https://github.com/pengzy-cosmo/DeepInflation.

DeepInflation: an AI agent for research and model discovery of inflation

TL;DR

DeepInflation addresses the challenge of exploring the expansive inflationary model space by fusing LLM-based reasoning, symbolic regression, and a retrieval-augmented knowledge base grounded in the Encyclopaedia Inflationaris. The approach autonomously discovers simple single-field slow-roll potentials that satisfy observational constraints on and (e.g., ACT DR6) and provides theoretical context for a broad set of inflationary scenarios. It demonstrates capabilities for both model discovery and literature-informed model inquiries, enabling researchers and non-experts to navigate the inflationary landscape through natural language. The work outlines a path toward autonomous scientific discovery in cosmology, while noting limitations to single-field models and outlining future extensions to multi-field dynamics, PBH production, and fully autonomous closed-loop operation.

Abstract

We present \textbf{DeepInflation}, an AI agent designed for research and model discovery in inflationary cosmology. Built upon a multi-agent architecture, \textbf{DeepInflation} integrates Large Language Models (LLMs) with a symbolic regression (SR) engine and a retrieval-augmented generation (RAG) knowledge base. This framework enables the agent to automatically explore and verify the vast landscape of inflationary potentials while grounding its outputs in established theoretical literature. We demonstrate that \textbf{DeepInflation} can successfully discover simple and viable single-field slow-roll inflationary potentials consistent with the latest observations (here ACT DR6 results as example) or any given and , and provide accurate theoretical context for obscure inflationary scenarios. \textbf{DeepInflation} serves as a prototype for a new generation of autonomous scientific discovery engines in cosmology, which enables researchers and non-experts alike to explore the inflationary landscape using natural language. This agent is available at https://github.com/pengzy-cosmo/DeepInflation.
Paper Structure (9 sections, 2 equations, 2 figures)

This paper contains 9 sections, 2 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Architecture
  3. Symbolic Regression
  4. RAG Knowledge Base
  5. Analysis and Visualization Tools
  6. Examples
  7. Example: Model Discovery (according to ACT DR6 results)
  8. Example: Model Inquiry
  9. Discussion and Outlook

Figures (2)

  • Figure 1: System architecture of DeepInflation. The system is composed of an Agno Agent Team and a SR engine. The Main Agent orchestrates the workflow, retrieving theoretical context from the Knowledge Base and delegating model discovery tasks to the SR Sub-Agent. The SR engine employs PySR for genetic search, coupled with a Julia-based Physics Kernel that evaluates the $\chi^2$ loss by solving the inflationary equations of motion.
  • Figure 2: Diagnostic plot generated by the agent for the discovered best model $V(\phi)=\exp\!\left(-\frac{0.42214}{\phi}\right)$.