Toward a Community Roadmap for High Energy Physics and Artificial Intelligence in China and Beyond

This paper presents a community-informed overview and initial roadmap for the development of Artificial Intelligence in High Energy Physics, drawing on discussions from the 2025 Qingdao workshop to guide future coordinated efforts in China and globally.

The Gist

The Big Picture: Two Giants Meeting

Imagine High Energy Physics (HEP) as a massive, high-stakes treasure hunt. Scientists are looking for tiny, rare clues (new particles) hidden inside mountains of data generated by giant machines like particle colliders.

Now, imagine Artificial Intelligence (AI) as a super-smart, tireless assistant that is incredibly good at finding patterns in chaos.

This paper is a "roadmap" written by scientists in China and their global partners. It says: "We have these two giants meeting, and it's going to be huge. But right now, everyone is working in their own little rooms. We need to build a bigger house where they can work together efficiently."

Why They Need Each Other

• Physics needs AI: The data from these experiments is so huge and messy (like trying to find a specific grain of sand on a beach while a hurricane is blowing) that humans can't sort it all out alone. AI helps filter the noise, spot the rare clues, and simulate what should happen.
• AI needs Physics: Physics problems are like a gym for AI. They are strict, logical, and follow exact rules (like gravity or math). By teaching AI to solve these hard, structured puzzles, scientists can make AI smarter, more reliable, and better at reasoning, not just guessing.

The Four Main Areas of Work

The paper breaks down how they are working together into four "rooms":

1. The Experiment Room (The Data Factory):

   • What they do: This is where the big machines (like the ones in China's JUNO or future CEPC projects) collect data.
   • The AI Role: AI acts like a super-fast security guard at the entrance. It decides in a split second which data is worth keeping and which is trash. It also helps design the machines themselves, making sure the hardware and the software work together perfectly from day one.

2. The Phenomenology Room (The Translator):

   • What they do: This group translates the raw data from the machines into actual physics stories.
   • The AI Role: Think of this as a translator who speaks both "Machine Code" and "Human Physics." They are trying to build a "Universal Translator" (called a Foundation Model) that learns the rules of physics once and can then help with many different types of experiments, rather than needing a new translator for every single job.

3. The Theory Room (The Mathematician):

   • What they do: These are the people who do the heavy math and theory before any data is even collected.
   • The AI Role: Instead of just crunching numbers, AI is becoming a "thinking partner." It helps explore complex mathematical landscapes to find solutions that are too hard for humans to calculate alone. It's like giving a mathematician a GPS that can navigate through a maze of equations to find the exit.

4. The Tool Room (The General Assistant):

   • What they do: This is about building general tools, like AI Agents (smart bots) and Large Language Models (chatbots trained on physics).
   • The AI Role: Imagine a personal assistant that knows every physics paper ever written. It can help a scientist write code, read a 50-page report in seconds, or set up a simulation. The goal isn't to replace the scientist, but to handle the boring, repetitive paperwork so the scientist can focus on the big ideas.

The Current Problem: Too Many Silos

The paper points out a major issue: Everyone is working alone.

• In China, different universities and labs have their own AI groups, but they aren't talking to each other enough.
• It's like having a dozen chefs in the same kitchen, each cooking a different dish, but none sharing ingredients or recipes.
• They lack a shared "pantry" (standardized data) and a shared "kitchen" (computing power).

The Proposed Solution: Building a Shared Kitchen

To fix this, the authors propose a "Community Roadmap" with three main pillars:

1. Shared Data (The Pantry):
   They want to create a central library where data is organized and ready for AI to use. Right now, data is often locked in messy formats that AI can't read. They want to standardize this so anyone can grab a dataset and start working immediately.

2. Shared Computing (The Kitchen):
   AI needs massive computing power (like giant ovens). The paper argues that smaller labs can't afford these ovens. They propose building a shared network where researchers across China can access powerful computers, similar to how everyone shares electricity from the same grid.

3. Training the Next Generation (The Apprentices):
   They need more people who speak both "Physics" and "AI." Currently, a physicist might know math but not coding, and a coder might know AI but not physics. The roadmap suggests creating special training programs, summer schools, and joint degrees to create "bilingual" scientists who can bridge the gap.

The Green Note

The paper also mentions that running these giant AI models uses a lot of electricity. They want to make sure this research is "Green AI"—using energy efficiently and not wasting resources, just as physicists try to be efficient with their experiments.

The Next Step: A Community Survey

The paper ends by saying, "We have a plan, but we need to check if everyone agrees."

• They launched a survey in late 2025 to ask scientists what they really need.
• They want to hear from everyone: students, professors, and industry experts.
• The goal is to gather enough feedback to write a bigger, more official "White Paper" that will guide funding and strategy for the next decade.

Summary

In short, this paper is a call to action. It says that the marriage between High Energy Physics and AI is already happening and is very powerful. However, to get the most out of it, the Chinese scientific community (and the world) needs to stop working in isolation, share their tools and data, and train a new generation of scientists who are experts in both fields.

Technical Summary

Technical Summary: Toward a Community Roadmap for High Energy Physics and Artificial Intelligence in China and Beyond

Problem Statement
While Artificial Intelligence (AI) has become a transformative force in data-intensive disciplines like biology and materials science, its integration into High Energy Physics (HEP) remains fragmented. Although the intersection of AI and HEP has a long history dating back to neural networks in the late 20th century, current efforts in China and the broader East Asian region are largely separated by institutional boundaries and lack a coherent, community-wide perspective. There is a critical need to articulate a unified roadmap that addresses the transition from exploratory, task-specific applications to a systematic, coordinated "AI-native" scientific ecosystem. Furthermore, while global initiatives (e.g., NSF's IAIFI, EuCAIF) have advanced cross-disciplinary integration, the specific landscape, infrastructure gaps, and strategic priorities within the Chinese HEP community require a dedicated overview to inform future coordinated efforts.

Methodology
This document functions as a community-informed overview rather than a primary experimental or theoretical study. The methodology involves:

1. Synthesis of Discussions: The content is motivated by and synthesizes discussions from the 2025 Quantum Computing and Machine Learning Workshop in Qingdao, China.
2. Literature and Activity Review: The authors review current AI activities across three core HEP domains: Experimental HEP (HEP-ex), Phenomenology (HEP-ph), and Theory (HEP-th), alongside emerging general-purpose AI tools.
3. Community Survey Integration: The document serves as a precursor to a comprehensive white paper, informed by an ongoing community survey launched in November 2025. This survey targets researchers across subfields, career stages, and institutions to gather data on demographics, technical focus, resource constraints, and infrastructure needs.
4. Categorization: The authors structure the landscape into four overlapping categories to map the current state of the field and identify emerging needs.

Key Contributions and Technical Landscape
The paper categorizes AI+HEP research into four primary domains, detailing specific technical applications and future directions:

• AI in High Energy Experiment (HEP-ex):

  • Current State: Deep learning is deeply integrated into workflows for particle identification, track reconstruction (using CNNs and Graph Neural Networks), and fast detector simulation (using generative models like flow-based models). Hardware trigger systems utilize boosted decision trees and compact neural networks for low-latency decisions.
  • Future Direction: The paper advocates for "co-design" in next-generation facilities (e.g., STCF, CEPC), integrating AI at the design stage rather than retrofitting. This includes embedding AI in trigger systems and edge-computing architectures (FPGAs/GPUs) for real-time anomaly detection and intelligent data selection.

• AI in High Energy Phenomenology (HEP-ph):

  • Current State: The field has moved from high-level observables to particle-level representations, utilizing task-specific models for jet tagging, simulation-based inference, and density estimation.
  • Future Direction: A shift toward foundation models pre-trained on large collections of experimental and simulated data. These models aim to learn shared physical representations, reducing the need for repeated task-specific training and enabling transfer across experiments. Key challenges include data standardization, benchmarking, and uncertainty quantification.

• AI in High Energy Theory (HEP-th):

  • Current State: AI is applied to structured mathematical problems (scattering amplitudes, Feynman integrals, lattice configurations) where the challenge is navigating constrained abstract spaces rather than pattern recognition.
  • Future Direction: The use of Reinforcement Learning (RL) and LLM-based agents to iteratively explore solution spaces guided by physical consistency. The paper highlights a bidirectional relationship where the rigorous demands of theoretical physics (exact reasoning, symbolic manipulation) are driving the development of neuro-symbolic methods and interpretable AI.

• General-Purpose AI Tools (LLMs and Agents):

  • Vision: Development of HEP-aware Large Language Models (LLMs) and agent-based systems to automate scientific workflows (literature review, code generation, hypothesis testing).
  • Role: These tools are envisioned not as replacements for researchers but as assistants for repetitive, technically detailed tasks, allowing scientists to focus on physical interpretation.
  • Challenges: Critical issues include reliability, hallucination control, reproducibility, and the integration of these tools with existing software frameworks.

Community Ecosystem and Infrastructure
The paper identifies significant gaps in the Chinese ecosystem compared to international counterparts:

• Infrastructure: There is a lack of standardized, machine-learning-ready datasets and uneven access to high-performance computing (HPC). The authors propose a unified open-data framework with dual-scale datasets (trial and full-scale), standardized metadata, and curated benchmarks.
• Computing: Calls for shared computing platforms, federated networks connecting institutional clusters, and strategic partnerships with cloud providers.
• Green AI: An emerging need to address the environmental impact and energy efficiency of large-scale AI training.
• Workforce: A gap exists in translating domain-specific physics needs into modern AI practice. The paper advocates for structured interdisciplinary training pathways, dual mentorship, and joint research programs between physics and computer science departments.

Results and Claims
As a roadmap document, the paper does not present experimental results or quantitative performance metrics. Instead, its "results" are the identification of the current state of the field and the articulation of priorities:

1. Transition Phase: The field is transitioning from community-driven, exploratory research to systematic, institutionally supported development.
2. Strategic Opportunity: China is well-positioned to integrate AI into the design of next-generation facilities (STCF, CEPC) from the outset, offering a unique strategic advantage.
3. Survey Status: As of the writing of this document, the accompanying community survey has received over 30 responses and aims to exceed 100 to generate statistically meaningful insights for a future white paper.

Significance
The paper positions itself as an initial roadmap and a "partial and evolving snapshot" rather than a definitive or complete representation of the field. Its significance lies in:

• Catalyzing Coordination: It aims to inform future coordinated efforts and lay the groundwork for a more comprehensive community white paper.
• Defining Priorities: It highlights the need for shared infrastructure (data, compute), standardized benchmarks, and workforce development to support the growing AI+HEP community in China.
• Global Context: It frames local developments within a broader global cooperative endeavor, emphasizing that HEP, with its culture of open collaboration, is well-positioned to foster a research culture where AI advances are guided by transparency, rigor, and scientific value.

The authors explicitly state that the perspectives presented are selective and reflect the views gathered from the authors and their collaborators, serving as a starting point for broader community input and strategic planning.