Results of the analysis of a survey for young scientists on training quality in HEP instrumentation software and machine learning

This report details the results of a survey conducted by the ECFA Early-Career Researchers Panel to assess the accessibility and quality of training programs in software and machine learning for instrumentation, aiming to guide improvements for early-career researchers in experimental physics.

The Gist

Imagine a massive, high-tech kitchen where thousands of chefs (scientists) are trying to cook the most complex dishes in the universe. These dishes are made of data from particle collisions. To cook them, the chefs need special tools: software to chop the data, machine learning to taste-test it, and electronics to keep the stove from exploding.

Recently, a group of young chefs (Early-Career Researchers) decided to ask a simple question: "Are we being taught how to use these tools properly?"

Here is the story of what they found, explained simply.

The Big Problem: "We Have the Tools, But No Manual"

A few years ago, it was discovered that 7 out of 10 young scientists were using open-source software (like free, community-built tools) to do their work. But here's the kicker: 7 out of 10 of them had never been taught how to use them.

It's like being handed a Ferrari and told, "Good luck, drive!" without ever sitting in a driver's seat or learning what the pedals do.

The Survey: Asking the Chefs

To fix this, a group called the ECFA Panel sent out a survey to 174 young scientists. They wanted to know:

1. Do you know where to find training?
2. Have you ever taken a class?
3. What do you actually want to learn?

Who answered?
Most were from Europe, working on giant projects like the Large Hadron Collider (LHC). They ranged from students just starting out to professors. Most of them spent their days analyzing data or designing detectors.

The Four Main "Cooking Classes"

The survey looked at four specific areas where scientists need training. Here is what they found in each:

1. Machine Learning (The "Smart Assistant")

• The Situation: Almost everyone (98%) uses or wants to use AI and Machine Learning. It's the hottest topic.
• The Problem: Most people are teaching themselves by reading manuals or asking a friend. Only a tiny fraction (9%) went to a dedicated school.
• What they want: They don't want long, boring lectures full of complex math. They want hands-on practice. Imagine a cooking class where you don't just watch the chef; you get to chop the onions and taste the sauce immediately. They want "best practices" and real-world examples, not just theory.

2. Detector Simulation (The "Virtual Reality Simulator")

• The Situation: This is like using a flight simulator before flying a real plane. About half the scientists use these tools.
• The Problem: Many don't know these schools exist. If they do know, they often feel the classes are too short or too theoretical.
• What they want: They want short, focused workshops and clear documentation (like a recipe book with pictures). They want to know exactly how to use the software, not just the history of why it was invented.

3. Data Acquisition & Control (The "Traffic Cop")

• The Situation: This is the system that catches the data and keeps the machine safe. About 43% of scientists are interested in this.
• The Problem: Many are using custom or commercial tools but feel they are flying blind. They don't know where to find training.
• What they want: Again, short workshops and clear guides. They want to learn how to fix the "traffic jams" in the data flow, not just listen to a lecture about traffic laws.

4. Detector Electronics (The "Wiring and Circuits")

• The Situation: This is the physical wiring that turns particle hits into digital signals. Only about 1/3 of the scientists are involved here.
• The Problem: This is the most neglected area. Zero respondents said they had ever attended a school specifically for this. Most are learning by trial and error or asking a senior colleague.
• What they want: They need comprehensive documentation and short, practical sessions. They want to understand how to build the circuits, not just read about them.

The Common Thread: What Everyone Agreed On

Across all four areas, the young scientists gave a very clear message:

1. Stop the "Lecture Hall" approach: They are tired of long, boring classes where the teacher talks for hours.
2. Give us the "Hands-On" experience: They want to get their hands dirty, make mistakes, and fix them with an expert nearby.
3. Make the "Menu" visible: The biggest complaint was that they didn't even know these classes existed. There is no central "Google" for these training programs.
4. Short and Sweet: They prefer short, frequent lessons (like snack-sized training) rather than week-long marathons.

The Solution: A New Recipe

The authors of the report suggest a few simple fixes to help the next generation of scientists:

• Create a Central Hub: Build a single website where you can easily find all training courses, filtered by topic and difficulty level. No more hunting through scattered emails.
• Publish the Manuals: Even if you can't attend a class, the notes, code, and videos from those classes should be available online for everyone to use.
• Focus on Practice: Design courses where 80% of the time is spent doing the work, and only 20% is spent listening to theory.

In a nutshell: The young scientists are eager to learn, but the current training system is like a library with no signposts and books written in a foreign language. They are asking for a friendly guide, a clear map, and a chance to practice what they learn.

Technical Summary

1. Problem Statement

High Energy Physics (HEP) instrumentation relies heavily on open-source software, machine learning (ML), and complex hardware systems. A 2021 study revealed a critical gap: while 71% of early-career researchers (ECRs) used open-source tools, 70% received no formal training for them.

The ECFA Early-Career Researchers (ECR) Panel identified a lack of accessible, high-quality, and structured training programs in four key domains:

1. Machine Learning (ML)
2. Detector Simulation
3. Data Acquisition (DAQ) and Detector Control Systems (DCS)
4. Detector Electronics (EDA)

The core problem is that current training is often too theoretical, lacks hands-on relevance to HEP analysis, suffers from poor visibility, and relies heavily on self-teaching or ad-hoc mentorship, leaving many ECRs under-equipped for modern experimental challenges.

2. Methodology

• Survey Design: A comprehensive survey was launched in 2025 by the "Software and Machine Learning for Instrumentation" group within the ECFA ECR Panel.
• Target Audience: Early-career researchers (postdocs, students, technical staff) in experimental and applied physics.
• Sample Size: 174 participants.
  • Geography: ~86% Europe, 14% Asia, Americas, and Africa.
  • Affiliation: Majority involved in Large Hadron Collider (LHC) collaborations (>1,000 members); others in fixed-target experiments (SPS), neutrino experiments (DUNE, T2K), and future projects (FCC, ILD).
• Structure:
  • General Section: Demographics, career stage, and general awareness of training.
  • Specialized Modules: Participants self-selected into four specific tracks (ML, Simulation, DAQ/DCS, Electronics) to answer detailed questions regarding usage, learning methods, and curriculum preferences.
• Data Analysis: Quantitative analysis of usage statistics and satisfaction levels, combined with qualitative analysis of open-ended feedback regarding training gaps.

3. Key Contributions

The paper provides the first granular, data-driven assessment of the training landscape for HEP instrumentation tools among ECRs. Its primary contributions include:

• Identification of the "Self-Taught" Dominance: Quantifying that the majority of skills acquisition occurs via self-study or peer mentorship rather than formal schools.
• Curriculum Design Guidelines: Establishing a consensus-based "ideal curriculum" structure (e.g., time allocation percentages) for future schools across all four domains.
• Gap Analysis: Highlighting specific deficits in current training, such as the lack of HEP-specific examples, insufficient hands-on practice, and poor discoverability of courses.
• Strategic Recommendations: Proposing concrete solutions for the community, including centralized course directories and the public release of training materials.

4. Key Results

A. General Findings

• Awareness Gap: Over 50% of respondents are unaware of available training opportunities.
• Participation: Only 28.7% have attended a relevant school.
• Satisfaction: Among attendees, >72% reported high satisfaction (>70%), but cited issues with pacing (too fast/advanced) and lack of practical application.
• Top Interests: ML and advanced statistical techniques are the highest demand areas.

B. Domain-Specific Results

| Domain | Usage/Interest | Primary Learning Method | Key Pain Points | Preferred Training Format |
| :--- | :--- | :--- | :--- :--- |
| Machine Learning (ML) | 98% use or are interested; 90% want to learn more. | Self-taught / Colleagues (66%). Only 9% via schools. | Lack of HEP-specific examples; math-heavy without application. | Comprehensive docs (25%), Workshops (21%), Online courses (32%). |
| Detector Simulation | 56% have experience; 6% do not need it. | Self-taught / Colleagues (80%). | 42% unaware of schools; 20% planning to attend. | Comprehensive docs (30%), Short workshops (24%). |
| DAQ / DCS | 43% involved/interested. | Self-taught / Colleagues (73%). | 47% unaware of schools; funding barriers (14%). | Docs & Short workshops (27% each). |
| Detector Electronics | 33% involved/interested. | Self-taught / Colleagues (83%). | 0% have attended a dedicated school. | Docs & Short workshops (30% each). |

C. Consensus on Curriculum Structure

Across all four domains, respondents agreed on a preferred time allocation for future schools:

• Hands-on sessions with experts: ~20%
• Workshops on best practices: ~20%
• Specific software sessions: ~15%
• Team/Individual development projects: ~15%
• Theoretical foundations: ~10% (Lowest priority)

5. Significance and Recommendations

This report serves as a foundational document for the ECFA and CERN to restructure HEP education. The significance lies in shifting the paradigm from "theoretical lectures" to "practical, HEP-centric application."

Key Recommendations:

1. Centralized Visibility: Create a centralized website with filters for course level and topic to solve the "awareness gap."
2. Documentation as Training: Make comprehensive documentation with HEP-specific examples the primary training deliverable, as it is the most requested format.
3. Format Shift: Move away from long, intensive schools toward short, focused workshops and online modules that allow for iterative learning.
4. Content Relevance: Ensure all training materials include direct applications to HEP analysis (e.g., jet tagging, event reconstruction) rather than generic software tutorials.
5. Resource Accessibility: Ensure training materials are publicly available for offline, self-paced learning to mitigate time-zone and funding barriers.

Conclusion: The HEP community is highly motivated to learn instrumentation software but is currently underserved by existing training infrastructure. By aligning future programs with the "hands-on, documentation-heavy, and short-format" preferences identified in this survey, the ECFA can significantly improve the technical readiness of the next generation of physicists.