(English)

Research Vision

Read the state of materials. Decide the next experiment.

AI⁴M Lab studies how energy materials are made, what internal states they actually have, how they change during use, and how these insights can guide the next experiment.

We connect experiments, computation, scientific data, physics, AI, and automation to build a human-guided closed-loop research system. AI and automation support repetitive measurement, analysis, and control, while researchers lead the scientific questions, interpretation, and final decisions.

<aside> 🎯

Priority Recruitment Areas

We are currently prioritizing researchers interested in scientific data platforms, LLM-based research systems, and AI modeling for experimental science. Prior experience in materials science is welcome but not required. Applicants from computer science, data science, statistics, applied mathematics, industrial engineering, and related fields are strongly encouraged to apply if they want to solve real scientific problems.

</aside>

Three Questions We Study

• Synthesizability — How can we make the intended state? We study how processing conditions and pathways determine the formation and transformation of materials, with an emphasis on reproducibility.
• Heterogeneity — What state does the material actually have? We combine multiple measurements to quantify local differences, boundaries, defects, and regions that cannot yet be confidently explained.
• Lifetime — How does the current state shape future performance? We combine experimental data with physical and data-driven models to forecast performance changes and degradation together with uncertainty.

How Our Research Connects

Experimental and process data → Material-state quantification → Physics and AI modeling → Performance and lifetime forecasting → Selection and execution of the next experiment

Each study is designed to become an input to the next. Short-term results are retained as reusable datasets, analysis methods, simulators, or experimental modules rather than isolated demonstrations.

Priority Research Areas

1. Scientific Data Platform

Connect heterogeneous experimental, computational, imaging, and analysis data through shared schemas and identifiers. Develop methods to track metadata, provenance, processing history, uncertainty, and version changes so that results can be reproduced and reused.

2. LLM-based Research System

Develop LLM systems that go beyond answering questions. The system should select and execute appropriate scientific tools, connect data and models, preserve the reasoning trail and evidence, and recover transparently from failures.

3. AI Modeling for Experimental Science

Develop reliable models for small, noisy, heterogeneous, and multimodal datasets. Research includes anomaly detection, relationship discovery, prediction, model selection, uncertainty estimation, and validation on unseen data.

4. Multimodal Quantification

Transform images, diffraction, spectroscopy, electrochemical, and process measurements into quantitative descriptions of material state while retaining unknown or low-confidence regions.

5. Physics-informed Simulation and Closed-loop Experimentation

Combine physical understanding with data-driven models to forecast material behavior, then connect model outputs to sensors, instruments, and software that support the selection and execution of the next experiment.

Student Entry Points

Undergraduate researchers can begin with well-bounded tasks such as:

• Organizing, visualizing, and quality-checking research data
• Designing metadata and simple data schemas
• Reproducing a baseline analysis or simulation
• Evaluating LLM outputs, tool selection, and failure cases
• Building a small data pipeline, API, sensor interface, or automated test

Research Culture

• Research question first: Scientific questions, rather than project names or fashionable tools, define the direction of the work.
• Quantitative and reproducible: We document provenance, uncertainty, negative results, and failure conditions as well as successful outcomes.
• Connected research: Individual results are designed to become reusable inputs for another experiment, analysis, or model.
• Human–AI collaboration: AI-generated outputs are critically evaluated against evidence, physical constraints, and independent validation.
• Models must be tested: Predictive accuracy alone is insufficient; we check data leakage, robustness, uncertainty, and performance on unseen data.

Who We Are Looking For

We welcome applicants from materials science, chemistry, chemical engineering, physics, computer science, data science, statistics, applied mathematics, industrial engineering, mechanical engineering, electrical engineering, control, robotics, and related fields.

You do not need experience in experiments, computation, AI, and automation all at once. A solid foundation in one or more of the following areas and a willingness to learn across boundaries are sufficient.

Preferred Experience

• Python-based data analysis or scientific computing
• Databases, APIs, data pipelines, or research software
• Data schemas, metadata, provenance, or version management
• Machine Learning development and statistical validation
• Small, noisy, or multimodal data analysis
• LLM tool use, agent workflows, RAG, or workflow orchestration
• Model evaluation, uncertainty estimation, or data-leakage prevention
• Automated testing and reproducible computational environments

Eligibility

• Prospective graduates or recent graduates.

💡 Note: Our lab covers a wide range of fields including experiments, computation, and coding. You do not need to be perfect in all these areas at the start. Even if you lack experience in a specific field, a passion for learning and utilizing new tools (AI) is all that matters.

Benefits: Why AI⁴M?

🎯 We provide top-tier support and environment so you can fully immerse yourself in research.

• Research Environment: We offer an exceptionally spacious and comfortable individual workspace. To maximize research efficiency, high-end workstations/PCs are provided to every researcher.
• AI Tools: Premium LLM (e.g., Claude Max, ChatGPT Pro) are provided to every researcher, enabling an AI-powered research workflow for maximum productivity.
• Financial Support: Full tuition coverage, monthly stipend, and performance-based incentives.
• Growth & Networking: On-campus housing (dormitory) available; active support for attending domestic and international conferences.

How to Apply

We welcome inquiries regarding admissions and lab life. If you are interested in applying, please send the following documents by email.

• Personal Statement (or Cover Letter)
• Curriculum Vitae (CV)
• Contact: [email protected]

Recruitment Schedule

• Graduate Students: We are recruiting for the Spring 2027 semester. Please contact us in advance according to the admission schedule. (📌 [DGIST Admission Guide])
• UGRP (Undergraduate Research): Undergraduates interested in AI-based material research are welcome to apply at any time.

(Korean)

연구 비전

소재의 상태를 읽고, 다음 실험을 결정한다.

AI⁴M Lab은 단순히 성능이 좋은 소재를 찾는 것을 넘어, 소재가 어떤 조건에서 만들어지고, 실제 내부 상태가 어떠하며, 사용하면서 어떻게 변하는지를 이해하는 연구를 수행합니다.

실험·계산·과학 데이터·물리·AI·자동화를 연결하여 연구자가 주도하는 폐루프 연구시스템을 구축하고 있습니다. AI와 자동화는 반복적인 측정·분석·제어를 돕지만, 연구 질문을 세우고 결과를 해석하며 최종 판단을 내리는 일은 연구자가 주도합니다.

<aside> 🎯

현재 우선 모집 분야

AI⁴M Lab은 소재 연구 데이터를 체계적으로 연결하는 Scientific Data Platform, 분석 도구와 모델을 연결하는 LLM-based Research System, 그리고 실제 실험 데이터에서 신뢰할 수 있는 결론을 도출하는 AI Modeling 분야의 연구자를 우선 모집합니다. 소재 연구 경험이 없더라도 데이터·AI·소프트웨어 역량을 실제 과학 문제 해결에 적용하고 싶은 분을 환영합니다.

</aside>

우리가 풀고자 하는 세 가지 질문

• 원하는 상태를 어떻게 만들 수 있는가? 공정 조건과 경로에 따라 소재가 어떻게 형성되고 변하는지 이해하고, 원하는 상태를 재현성 있게 만드는 방법을 찾습니다.
• 실제 소재는 어떤 상태인가? 여러 측정 결과를 함께 분석하여 소재 내부의 차이, 경계, 결함과 아직 확실히 설명되지 않은 영역을 정량화합니다.
• 현재 상태가 미래 성능에 어떤 영향을 주는가? 실험 데이터와 물리·데이터 기반 모델을 결합하여 성능 변화와 열화를 불확실성과 함께 예측합니다.

연구가 연결되는 방식

실험·공정 데이터 → 소재 상태 정량화 → 물리·AI 모델링 → 성능과 수명 예측 → 다음 실험 제안 및 실행

각 연구는 독립적으로 끝나지 않습니다. 실험에서 얻은 데이터는 분석과 모델의 입력이 되고, 모델의 결과는 다음에 수행할 실험이나 측정을 결정하는 근거가 됩니다. 단기 연구 결과도 재사용 가능한 데이터, 분석 방법, 시뮬레이터 또는 실험 모듈로 축적합니다.

우선 연구 분야

1. Scientific Data Platform

서로 다른 실험·계산·이미지·분석 데이터를 공통 schema와 ID로 연결합니다. 데이터의 metadata, 출처, 처리 과정, 불확실성과 version을 추적하여 결과를 재현하고 새로운 연구에 다시 사용할 수 있는 방법을 개발합니다.

2. LLM-based Research System

질문에 답하는 데 그치지 않고, LLM이 필요한 과학 데이터와 분석 도구를 선택·실행하고 모델을 연결하도록 합니다. 판단의 근거와 처리 과정을 남기고, 실패 원인을 확인하고 복구할 수 있는 연구 workflow를 개발합니다.

3. AI Modeling for Experimental Science

작고, 불균일하며, noise가 많고, 여러 형태가 섞인 실험 데이터에서 신뢰할 수 있는 결론을 얻는 모델을 개발합니다. 이상 탐지, 관계 발견, 예측, 모델 선택, uncertainty estimation과 새로운 데이터에서의 검증을 다룹니다.

4. Multimodal Quantification

이미지, 회절, 분광, 전기화학 및 공정 측정 결과를 소재 상태에 대한 정량적인 정보로 변환합니다. 설명하기 어려운 영역을 억지로 분류하지 않고 unknown 또는 낮은 confidence로 남겨 후속 측정과 연결합니다.

5. Physics-informed Simulation and Closed-loop Experimentation

물리적 이해와 데이터 기반 모델을 결합하여 소재의 거동을 예측합니다. 모델 결과를 센서·장비·소프트웨어와 연결하여 다음 실험 조건을 선택하고 실행하는 연구 흐름을 구축합니다.

학부생이 시작할 수 있는 연구