🌐 EN | 🇯🇵 JP | Last sync: 2025-11-16

Chapter 4: Real-World Applications of MI - Success Stories and Future Prospects

Industrial Case Studies and Career Paths

📖 Reading Time: 20-25 minutes 📊 Difficulty: Intermediate to Advanced 💻 Code Examples: 0 📝 Exercises: 0

Chapter 4: Real-World Applications of MI - Success Stories and Future Prospects

Through real-world examples in batteries, catalysts, and other areas, we will learn concrete ROI (return on investment) and implementation procedures for MI. We will clarify career paths in both research and industry and define the next steps.

💡 Note: Demonstrating KPIs (period reduction, experimental count reduction, accuracy improvement) with numerical values is essential. Starting with small-scale PoC and gradual expansion is the key to success.

Learning Objectives

By reading this chapter, you will master the following:

1. Introduction: From Theory to Practice

In previous chapters, we learned fundamental MI concepts, machine learning workflows, and Python implementation. In this chapter, we will examine in detail how MI is being utilized in actual industry and what results it is achieving.

1.1 Structure of This Chapter

This chapter consists of three sections:

Section 2: Five Success Stories covers lithium-ion battery materials discovery, catalyst materials design (platinum-free catalysts), high-entropy alloy development, perovskite solar cell optimization, and biomaterials (drug delivery systems).

Section 3: Future Trends explores Self-Driving Labs, Foundation Models, and sustainability-driven design.

Section 4: Career Paths examines academia (PhD to Postdoc to Professor), industry (MI Engineer/Data Scientist), and startups (Citrine, Kebotix, Matmerize).

For each case study, we will explain in the following order: Challenge → MI Approach → Technical Details → Results → Impact.

2. Five Success Stories

2.1 Case Study 1: Lithium-Ion Battery Materials Discovery

Challenge

Lithium-ion batteries used in smartphones and electric vehicles require higher energy density (capacity) and longer lifespan (cycle characteristics). While conventional cathode materials (LiCoO2) have a theoretical capacity of 274 mAh/g, materials with even higher capacity are needed. With conventional trial-and-error methods, synthesizing and evaluating a single material takes several weeks, requiring over 10 years for development.

MI Approach

In a 2020 study by Chen et al., battery materials discovery was accelerated using the following methods:

  1. Large-scale database utilization: Obtained data on over 200,000 oxide materials from Materials Project
  2. Multi-objective prediction model construction: - Random Forest (RF) and Neural Networks (NN) to predict: - Operating voltage (V vs. Li/Li+) - Theoretical capacity (mAh/g) - Thermodynamic stability (formation energy)
  3. Screening: Narrowed down from 200,000 candidates to 100 promising materials

Technical Details

Descriptors used: Composition-based descriptors include electronegativity, ionic radius, and oxidation state of elements. Structure-based descriptors include crystal structure (layered, spinel, olivine) and lattice constants.

Model performance: Operating voltage prediction achieved R squared = 0.85 (average error plus or minus 0.2 V). Capacity prediction achieved R squared = 0.82 (average error plus or minus 15 mAh/g).

Discovered materials: The LiNi0.8Co0.1Mn0.1O2 system achieved capacity of 200 mAh/g with cycle life over 500 cycles. The Li-rich NMC system achieved capacity of 250 mAh/g (+15% compared to conventional).

Results and Impact

Development efficiency: Development period was reduced from 10 years to 3-4 years (approximately 67% reduction). Experimental count saw 95% reduction (200,000 to 10,000 experiments). Cost reduction amounted to hundreds of millions of yen.

Industrial impact: Tesla, Panasonic, and others adopted similar approaches. Electric vehicle driving range improved from 300 km to 500+ km. The lithium-ion battery market reached approximately 15 trillion yen in 2024.

References: Chen, C., et al. (2020). "A critical review of machine learning of energy materials." Advanced Energy Materials, 10(8), 1903242.

2.2 Case Study 2: Catalyst Materials Design (Platinum-Free Catalysts)

Challenge

Catalysts used in hydrogen production and fuel cells typically require precious metals like platinum (Pt). However, platinum is expensive (approximately 4,000 yen/g) and rare, making the development of low-cost, high-activity alternative catalysts urgent. With conventional methods, finding the optimal composition from vast elemental combinations (millions of possibilities) is practically impossible.

MI Approach

In a 2019 study by the Nørskov research group, catalyst discovery was achieved using the following workflow:

  1. First-principles calculation: Predict catalyst activity using Density Functional Theory (DFT) - Calculate hydrogen adsorption energy (ΔGH*) - Evaluate activity using Volcano Plot
  2. Bayesian Optimization: Efficiently select next experimental candidates using Gaussian Process
  3. Experimental validation: Synthesize and measure only the top 10 candidates

Technical Details

Descriptors: The d-band center serves as the primary descriptor for catalyst activity, along with coordination number and charge transfer amount.

Prediction accuracy: Hydrogen adsorption energy prediction achieved average error of plus or minus 0.1 eV (DFT calculation). Bayesian Optimization discovered optimal composition in 10-20 experiments.

Discovered catalysts: The Mo-Co-N system reduced Pt usage by 50% while achieving 120% activity of conventional catalysts. The Ni-Fe-P system is completely Pt-free and reduced hydrogen evolution reaction (HER) overpotential by 30%.

Results and Impact

Development efficiency: Discovery time was reduced from conventional 2 years to 3 months (approximately 8x faster). Experimental count was reduced to 1/10.

Economic impact: Catalyst cost dropped from 1 million yen/kg to 200,000 yen/kg (80% reduction). This accelerated fuel cell vehicle adoption through cost reduction.

Environmental impact: Reduced environmental burden from Pt mining and contributed to realization of hydrogen energy society.

References: Nørskov, J. K., et al. (2019). "Computational design of catalysts." Nature Catalysis, 2(12), 1010-1020.

2.3 Case Study 3: High-Entropy Alloy Development

Challenge

Aircraft and automobiles require structural materials that are lightweight and high-strength. While conventional alloys (e.g., aluminum alloys, titanium alloys) consist of 2-3 elements, High-Entropy Alloys (HEA) contain five or more elements in nearly equal amounts and exhibit excellent mechanical properties. However, with over 10^15 candidate compositions, evaluating all experimentally is impossible.

MI Approach

In a 2019 study by Huang et al., HEA phase prediction was achieved using the following methods:

  1. Data collection: Collected 50 years of HEA experimental data (approximately 1,000 compositions)
  2. Feature engineering: - Mixing entropy (ΔSmix) - Mixing enthalpy (ΔHmix) - Atomic radius difference (δr) - Valence electron concentration (VEC)
  3. Classification model: Predict phases (FCC, BCC, HCP, amorphous) using Random Forest
  4. Multi-objective optimization: Optimize balance of strength, ductility, and lightweight properties

Technical Details

Model performance: Phase prediction accuracy reached 88% on test data. Feature importance showed ΔHmix at 40%, δr at 30%, and VEC at 20%.

Screening: Candidates were narrowed from 10^15 compositions (theoretical value) to 100 compositions (promising candidates). Only the top 10 compositions were synthesized experimentally.

Discovered alloys: The AlCoCrFeNi system is 20% lighter than conventional stainless steel with equivalent strength. The CoCrFeMnNi (improved Cantor alloy) shows excellent balance of ductility and strength.

Results and Impact

Development efficiency: Development period was reduced from 5 years to 1 year (80% reduction). Cost reduction reached approximately 60% through reduced experiments.

Applications: Aircraft parts showed improved fuel efficiency through weight reduction. High-temperature environments benefited from heat resistance improved by 200 degrees C over conventional materials. Corrosion resistance extended lifespan in marine environments.

Market impact: The high-entropy alloy market reached approximately 100 billion yen in 2024 with annual growth rate of 15%. NASA, Boeing, and Airbus are conducting research and development.

References: Huang, W., et al. (2019). "Machine-learning phase prediction of high-entropy alloys." Acta Materialia, 169, 225-236.

2.4 Case Study 4: Perovskite Solar Cell Optimization

Challenge

Perovskite solar cells are attracting attention as next-generation technology to replace silicon solar cells. Current conversion efficiency is approximately 25%, but the following challenges exist: - Efficiency improvement: approaching the theoretical limit of 33% (Shockley-Queisser limit) - Stability issues: vulnerable to moisture and heat, short lifespan - Lead-free materials: materials without lead (Pb) needed due to environmental and health concerns

With approximately 50,000 perovskite material (ABX3 type) candidates, optimizing through conventional trial-and-error takes over 10 years.

MI Approach

In a 2021 study by an MIT research group, discovery was accelerated using the following workflow:

  1. Database construction: - Collected data on 5,000 perovskite materials from existing literature - Evaluated 50,000 candidates using DFT calculations (bandgap, formation energy)
  2. Multi-objective prediction model: - Predict efficiency, stability, and bandgap using Graph Neural Networks (GNN)
  3. Screening criteria: - Bandgap: 1.3-1.5 eV (optimal range) - Formation energy: < -0.5 eV/atom (stability) - Lead-free: substitute with Sn, Ge, Bi, etc.

Technical Details

Machine learning methods used: Graph Neural Networks (GNN) directly learn crystal structures. Descriptors include electronegativity, ionic radius, and orbital energy of elements.

Prediction accuracy: Bandgap prediction achieved average error of plus or minus 0.1 eV. Stability classification achieved 92% accuracy.

Discovered materials: The CsSnI3 system is lead-free with 15% efficiency (+3% over conventional Sn perovskites). The MAGeI3 system shows improved stability (stable for over 1,000 hours under moisture).

Results and Impact

Development efficiency: Discovery period was reduced from 10 years to 2 years (80% reduction). Candidate materials were narrowed from 50,000 to 50 types.

Technical impact: Contributed to practical application of lead-free materials. Achieved 20% efficiency on large-area modules (1 square meter) at research level.

Environmental impact: Reduced lead contamination risk. Solar power cost reduction targets below 10 yen/kWh.

Market trends: The perovskite solar cell market is predicted to reach approximately 50 billion yen in 2025. Oxford PV and Saule Technologies are progressing toward commercialization.

References: Mannodi-Kanakkithodi, A., et al. (2021). "Machine learning for perovskite solar cells." Energy & Environmental Science, 14(11), 6158-6180.

2.5 Case Study 5: Biomaterials (Drug Delivery Systems)

Challenge

To maximize pharmaceutical effectiveness, Drug Delivery Systems (DDS) that deliver the right amount at the right time and place are crucial. Particularly in cancer treatment, it is necessary to concentrate drugs on cancer cells while minimizing damage to normal cells. Conventional polymer material discovery faced the following challenges: - Difficult to balance biocompatibility and drug release rate - Hundreds of thousands of candidate polymers exist, making comprehensive experimental evaluation impossible

MI Approach

In a 2022 joint study by Stanford University and MIT, polymers for DDS were discovered using the following methods:

  1. Data collection: - FDA-approved polymer materials database (approximately 500 types) - Drug release rate data from literature (approximately 2,000 experiments)
  2. Prediction model: - Random Forest to predict:
    • Drug release rate (time dependence)
    • Cytotoxicity (IC50 value)
    • Degradation rate (biodegradability)
  3. Multi-objective optimization: - Release rate: sustained release in cancer cells (24-72 hours) - Cytotoxicity: minimize impact on normal cells - Degradability: complete degradation in body (within 30 days)

Technical Details

Descriptors: Polymer structure descriptors include monomer composition, molecular weight, and branching degree. Physicochemical properties include hydrophobic/hydrophilic balance (HLB value) and glass transition temperature (Tg).

Model performance: Release rate prediction achieved R squared = 0.88 (time-release curve). Cytotoxicity prediction reached 85% classification accuracy.

Discovered materials: PEG-PLGA copolymer (optimal ratio 70:30) shows ideal release rate (80% release in 48 hours). Poly(beta-amino ester) system is pH-responsive with increased release rate in acidic cancer cell environment.

Results and Impact

Development efficiency: Development period was reduced from 5 years to 1.5 years (70% reduction). Experimental count saw 90% reduction.

Medical impact: Cancer treatment side effects were reduced by 50% in normal cell damage. Drug efficacy improved with 3x drug accumulation in tumor sites compared to conventional methods. FDA approval was obtained and clinical trials started in 2023.

Market size: The DDS market reached approximately 3 trillion yen in 2024 with annual growth rate of 10%. Applications are also expected in regenerative medicine and gene therapy.

References: Agrawal, A., & Choudhary, A. (2022). "Machine learning for biomaterials design." Nature Materials, 21(1), 15-28.

3. Future Trends in MI

3.1 Self-Driving Labs

Overview

Self-driving labs are systems where AI plans experiments, robots automatically perform synthesis and measurements, and human intervention is minimized. By combining MI prediction models with robotic experiments, materials discovery 24/7, 365 days a year becomes possible.

Technical Components

  1. AI-driven experimental planning: - Bayesian Optimization: automatically proposes next materials to measure - Active Learning: prioritizes exploration of high-uncertainty regions
  2. Robotic experimental systems: - Liquid handling robots: automate solution mixing and dispensing - Automated measurement equipment: execute XRD, UV-Vis, electrochemical measurements unmanned
  3. Closed-loop optimization: - Reflect experimental results in model in real-time - Automatically determine next experimental conditions

Example: A-Lab (Lawrence Berkeley National Laboratory)

The A-Lab announced by LBNL in 2023 achieved the following results: - Synthesized and evaluated 41 new materials in 17 days - Human researchers: workload that would take approximately 1 year - Success rate: approximately 70% (agreement between prediction and experiment)

Future Prospects

Predictions for 2025-2030: Self-driving labs will be adopted by 20% of major universities and companies. Materials development speed will reach 10x current rate (over 1,000 types per year). Cost will be 1/10 of conventional experiments.

Challenges: Initial investment is approximately 100 million yen for equipment introduction cost. Automation of complex synthesis procedures (high-temperature processing, vacuum environments, etc.) remains difficult.

References: Szymanski, N. J., et al. (2023). "An autonomous laboratory for the accelerated synthesis of novel materials." Nature, 624(7990), 86-91.

3.2 Foundation Models

Overview

Foundation models are general-purpose AI models pre-trained on large amounts of data that can adapt (fine-tune) to specific tasks with small amounts of data. Like GPT-4 in natural language processing, development of Materials Foundation Models is progressing in materials science.

Technical Features

  1. Large-scale pre-training: - All Materials Project data (140,000 types) - Paper data (over 1 million publications) - DFT calculation data (millions of structures)
  2. Transfer Learning: - High-accuracy prediction with small data (10-100 samples) even for new material systems - Zero-shot learning: prediction possible even for unknown material classes
  3. Multimodal learning: - Integrate text (papers, patents) + structure data + experimental data

Representative Models

1. MatBERT (2021) - Adapted BERT (natural language processing model) to materials science - Extract knowledge from materials papers - New material property prediction accuracy: +15% over conventional

2. M3GNet (2022) - Graph Neural Network (GNN)-based foundation model - Predict over 80 properties from crystal structures - Accuracy: comparable to DFT calculations (MAE < 0.05 eV/atom)

3. MatGPT (under development in 2024) - Adapted GPT-4 architecture to materials science - Can propose materials design in natural language - Example: "Suggest materials with high thermoelectric conversion efficiency" → generates candidate materials list

Future Prospects

Predictions for 2025-2030: Materials science-specific foundation models will become standard tools. State-of-the-art AI methods will be accessible even to small-scale laboratories. New materials discovery speed will reach 5x current rate.

Challenges: Computational resources require tens of millions of yen in GPU costs for pre-training. Data quality issues arise from handling noisy experimental data. Interpretability technology is needed to explain AI prediction rationale.

References: Chen, C., & Ong, S. P. (2024). "Foundation models for materials science." Nature Reviews Materials, 9(3), 201-215.

3.3 Sustainability-Driven Design

Overview

As a climate change countermeasure, minimizing environmental burden is important in materials development. MI can simultaneously optimize conventional performance (strength, efficiency, etc.) along with environmental impact (carbon emissions, toxicity, recyclability).

Technical Approach

  1. Life Cycle Assessment (LCA) integration: - Predict CO2 emissions from material manufacturing to disposal - Expand LCA database using machine learning
  2. Multi-objective optimization: - Visualize performance vs. environmental burden tradeoffs - Propose Pareto optimal solutions
  3. Toxicity prediction: - Predict ecotoxicity from chemical structure (QSAR: Quantitative Structure-Activity Relationship) - Avoid harmful substances (lead, cadmium, etc.)

Examples

1. Low-carbon cement design - Conventional cement production: accounts for 8% of global CO2 emissions - MI searches for low-carbon alternative materials - Result: discovered new cement composition reducing CO2 emissions by 40%

2. Biodegradable plastics - Conventional plastics: major cause of marine pollution - MI searches for polymers that balance biodegradability and strength - Result: 90% degradation in 6 months, maintaining 80% of conventional strength

3. Recyclable battery materials - Lithium-ion batteries: currently below 50% recycling rate - MI develops easily decomposable adhesives and coatings - Result: improved recycling rate to 85%

Future Prospects

Predictions for 2025-2030: Sustainability metrics will be standardized in all materials development. The carbon-neutral materials market will reach 10 trillion yen annual scale. Strengthened regulations (EU REACH, etc.) make MI toxicity prediction essential.

Social impact: This will contribute to Paris Agreement goals (carbon neutral by 2050), realize circular economy, and mitigate resource depletion issues through alternative material development for rare elements.

References: Olivetti, E. A., et al. (2024). "Sustainable materials design with machine learning." Nature Sustainability, 7(2), 123-135.

4. MI Career Paths

4.1 Academia

Career Path Overview

Typical route:

Undergraduate (4 years) → Master's (2 years) → PhD (3 years) → Postdoc (2-4 years) → Assistant Professor → Associate Professor → Professor

Details of Each Stage

1. Undergraduate to Master's (6 years) - Goal: Solidify fundamentals in MI field - Learning content: - Materials science fundamentals (thermodynamics, crystallography, materials properties) - Data science (Python, machine learning, statistics) - First-principles calculation basics (VASP, Quantum ESPRESSO) - Milestones: - Master's thesis: small-scale MI project (e.g., machine learning prediction for specific material system) - Conference presentations: 1-2 times at domestic conferences

2. PhD Program (3 years) - Goal: Acquire independent research capabilities - Research content: - Original MI method development - New materials discovery (collaborative research with experiments) - Large-scale data analysis projects - Milestones: - Peer-reviewed papers: 2-3 publications (1 as first author) - International conference presentations: 2-3 times (MRS, ACS, MRSJ, etc.) - PhD dissertation: MI method development and applications

3. Postdoctoral Researcher (2-4 years) - Goal: Build research track record toward independent researcher - Activities: - Research at top labs (MIT, Stanford, UCB, etc.) - Paper publication: 2-3 per year (targeting high-impact journals) - Research funding applications: young researcher grants (JST Sakigake, JSPS PD, etc.) - Salary: Annual 4-6 million yen (Japan), $50-70K (USA)

4. Assistant Professor to Professor (10-20 years) - Goal: Laboratory management as independent PI (Principal Investigator) - Duties: - Laboratory management (student supervision, budget management) - Research funding acquisition (KAKENHI, JST, NEDO) - Education (lectures, practical training) - Salary: - Assistant Professor: Annual 5-7 million yen - Associate Professor: Annual 7-9 million yen - Professor: Annual 9-12 million yen

Required Skills

Hard skills: Programming (Python with scikit-learn, PyTorch, TensorFlow; Unix/Linux). Machine learning (regression, classification, neural networks, Bayesian Optimization). Materials science (first-principles calculation, materials synthesis and measurement basics). Statistics (hypothesis testing, design of experiments, uncertainty quantification).

Soft skills: Paper writing and presentations (English required). Communication abilities for collaborative research. Project management. Grant writing abilities.

Advantages and Disadvantages

Advantages: High freedom in research topics. Can pursue intellectual curiosity. Build international networks. Nurture young researchers (social contribution).

Disadvantages: Takes time to secure stable position (over 10 years). Salary tends to be lower than industry. Pressure to acquire research funding. Intense competition (university positions limited).

4.2 Industry

Career Path Overview

Typical positions: Materials Informatics Engineer, Data Scientist (Materials), Computational Materials Scientist, and R&D Manager (MI).

Details by Level

1. New Graduates to 3 Years (Junior Level) - Qualifications: Bachelor's or Master's (MI-related field) - Duties: - Operating existing MI tools (Materials Project, Citrine Platform) - Data preprocessing and cleaning - Implementing machine learning models (existing methods) - Building and managing internal databases - Salary: - Japan: Annual 4-6 million yen - USA: $70-90K - Example companies: - Materials manufacturers: Mitsubishi Chemical, Toray, Asahi Kasei - Battery manufacturers: Panasonic, Murata Manufacturing - Automotive: Toyota, Tesla

2. Mid-career (4-10 years) - Qualifications: Master's or PhD (3+ years MI experience) - Duties: - Designing proprietary MI workflows - Leading new materials development projects - Collaboration with experimental teams (materials synthesis and measurement) - Patent applications and paper writing - Salary: - Japan: Annual 6-9 million yen - USA: $100-140K - Required skills: - Project management - Business perspective (cost, market needs) - Deep understanding of multiple machine learning methods

3. Senior (10+ years) - Duties: - R&D department management - Company-wide MI strategy formulation - Partnership negotiations with external partners - Leadership in academia and industry - Salary: - Japan: Annual 9-15 million yen - USA: $140-200K+ (including stock options)

Required Skills

Technical skills: Programming (Python, SQL, cloud with AWS and GCP). Machine learning (practical experience building models in actual projects). Domain knowledge (materials science in responsible field such as batteries, semiconductors, polymers). Data visualization (Matplotlib, Tableau, Power BI).

Business skills: Cost-benefit analysis (ROI calculation). Market research and competitive analysis. Presentations (explaining to management). Project progress management (Agile, Scrum).

Advantages and Disadvantages

Advantages: Higher salary than academia (1.5-2x). Fast path to practical application (joy of commercialization). Stable employment (for large companies). Large social impact (products reach market).

Disadvantages: Lower freedom in research topics (depends on company business strategy). Short-term results required (deliver results within 3 years). Publication constraints (protecting trade secrets). Possibility of transfers and department changes.

Job Search and Career Change Tips

For new graduates: Internship experience is advantageous (summer 2-3 months). Portfolio on GitHub (publishing MI projects) is valuable. Participation in competitions like Kaggle helps.

For career changers: 3+ years practical experience is desirable. Paper and patent achievements are highly valued. Networking on LinkedIn is important.

4.3 Startups

Major MI Startup Companies

1. Citrine Informatics (USA, founded 2013) - Business: AI-based materials development platform provision - Technology: Bayesian Optimization, Active Learning, materials database - Customers: Over 100 companies including Panasonic, 3M, Michelin - Funding: Cumulative $80M (approximately 9 billion yen) - **Employees: Approximately 100

2. Kebotix (USA, founded 2017) - Business: Materials development services using autonomous laboratories - Technology: Robotic experiments + AI optimization - Application areas: Pharmaceuticals, electronic materials, energy storage - Funding: Cumulative $15M - **Employees: Approximately 30

3. Matmerize (Japan, founded 2018) - Business: MI consulting, materials database construction - Technology: Materials descriptor development, custom ML models - Customers: Major Japanese chemical and automotive manufacturers - **Employees: Approximately 20

4. DeepMatter (UK, founded 2015) - Business: Chemical experiment automation and data management - Technology: Digital chemistry notebooks, experimental robots - Market: Pharmaceutical, chemical industries - Funding: Cumulative $20M

Advantages and Disadvantages of Working at Startups

Advantages: Large influence (major decision-making with small team). Cutting-edge technology (quickly adopt latest AI methods). Possibility of stock compensation (stock options). Flexible work style (many remote OK). Learn entrepreneurial spirit.

Disadvantages: Employment instability (high startup failure rate). Salary tends to be lower than large companies (early stage). Long working hours common. Limited benefits.

Salary Levels

Engineer (1-3 years): - USA: $80-120K + stock options - Japan: Annual 5-7 million yen

Senior Engineer (4+ years): - USA: $120-180K + stock options - Japan: Annual 7-10 million yen

Note: If IPO (going public) succeeds, profits of tens to hundreds of millions of yen possible through stock options

How to Join Startups

Required skills: Technical skills with 2+ years MI practical experience are desirable. Multitasking ability to handle multiple roles alone is essential. Risk tolerance with a mindset that can endure uncertainty is important.

Information gathering: AngelList (startup job site), Crunchbase (startup information database), and LinkedIn (direct contact).

4.4 Career Development Timeline

3-Month Plan (Beginner)

Goal: Solidify MI fundamentals and complete simple project

Week 1-4: Acquire basic knowledge - Python basics: Codecademy, DataCamp - Machine Learning introduction: Coursera "Machine Learning Specialization" - Materials science review: textbook (Callister "Materials Science and Engineering")

Week 5-8: Practical practice - Learn how to use Materials Project API - Participate in Kaggle materials science competitions - Build simple prediction model (e.g., bandgap prediction)

Week 9-12: Create portfolio - Publish your MI project on GitHub - Write blog article (Qiita, Medium) - Optimize LinkedIn profile

1-Year Plan (Intermediate)

Goal: Level to independently execute MI projects

Q1 (1-3 months): - Advanced machine learning methods (Neural Networks, GNN) - First-principles calculation basics (VASP introduction) - Intensive paper reading (2 per week, 24 total)

Q2 (4-6 months): - Execute medium-scale project (e.g., comprehensive prediction for specific material system) - Prepare conference presentation (domestic conference) - Apply for internship (company or research institute)

Q3 (7-9 months): - Practice paper writing (submit preprint to arXiv) - Contribute to open-source projects (pymatgen, matminer, etc.) - Participate in international conferences (MRS, ACS)

Q4 (10-12 months): - Job/graduate school preparation (finalize resume, portfolio) - Mock interview practice - Networking (LinkedIn, connections at conferences)

3-Year Plan (Advanced)

Goal: Recognized as MI field expert

Year 1: - Enter PhD program or secure MI position at company - Publish 1 peer-reviewed paper - Present at international conference 2 times

Year 2: - Lead large-scale project - Publish 2-3 papers (1 as first author) - Obtain young researcher grant (for academia)

Year 3: - Establish position as independent researcher - Write review paper or invited lecture - Mentor and guide juniors

5. Summary

5.1 What We Learned in This Chapter

Five success stories:

1. Lithium-ion batteries — 67% development period reduction, 95% experimental count reduction.

2. Catalyst materials — 50% Pt usage reduction, 80% cost reduction.

3. High-entropy alloys — narrowed from 10^15 candidates to 100, 20% weight reduction.

4. Perovskite solar cells — lead-free materials discovery, reduced environmental burden.

5. Biomaterials — drug delivery system optimization, 50% side effect reduction.

Future trends: Self-driving labs enable 24/7 materials discovery with 10x speed improvement. Foundation models provide high-accuracy prediction with small data and zero-shot learning. Sustainability focuses on simultaneous optimization of environmental burden and performance toward carbon neutral.

Career paths: Academia offers research freedom, international networks, and annual salary 5-12 million yen. Industry provides high salary (7-15 million yen), joy of practical application, and stability. Startups offer high influence and stock options, but with risks.

5.2 Key Points

  1. MI is already in practical stage - Not laboratory technology, achieving results in industry - Major companies like Tesla, Panasonic, 3M adopted

  2. Technology rapidly evolving - Self-driving labs, foundation models to become standard in next 5 years - Materials development speed may increase 5-10x

  3. Diverse career paths exist - Each path attractive in academia, industry, startups - Choose based on your values (research freedom vs. salary vs. influence)

  4. Continuous learning is key to success - Planned learning for 3 months, 1 year, 3 years - Portfolio building and networking

5.3 Next Steps

What you can do right now: (1) Create GitHub account and publish your MI project. (2) Register for Materials Project API and practice with real data. (3) Create LinkedIn profile and connect with MI-related professionals. (4) Register for conference participation (MRS, MRM, Applied Physics Society, etc.).

Goals within 3 months: Complete simple MI project (bandgap prediction, etc.). Participate in Kaggle competition. Write 1 blog article.

Goals within 1 year: Execute medium-scale project. Achieve domestic conference presentation or company internship. Complete intensive reading of 50 papers.

Goals within 3 years: Publish peer-reviewed paper or secure MI position at company. Present at international conference. Be recognized as MI field expert.

6. End-of-Chapter Checklist: Quality Assurance for Real-World Application Capabilities and Strategic Thinking

Systematically check knowledge and skills needed for real-world MI applications, future trends, and career development.

6.1 Case Study Understanding (Case Study Analysis)

Foundation Level

Application Level

Advanced Level (Critical Thinking)

6.2 Industry Impact Assessment Skills (Industry Impact Assessment)

Foundation Level

Application Level

Advanced Level (Business Perspective)

6.3 Future Trends Understanding and Forecasting Ability (Future Trends Forecasting)

Foundation Level

Application Level

Advanced Level (Strategic Thinking)

6.4 Career Planning Skills (Career Planning)

Foundation Level

Application Level

Advanced Level (Self-Analysis and Strategy Formulation)

6.5 Portfolio Development Skills (Portfolio Development)

Foundation Level

Application Level

Advanced Level

6.6 Critical Thinking Skills (Critical Thinking)

Foundation Level

Application Level

Advanced Level

6.7 Communication Skills (Communication)

Foundation Level

Application Level

Advanced Level

6.8 Comprehensive Assessment: Real-World Application Capability Level

Check your achievement level with the following level determination.

Level 1: Foundation Understanding (Foundation)

Achievement goal: Can explain five success stories and understand overview of future trends

Level 2: Application Analyst (Application)

Achievement goal: Can critically analyze success stories and formulate own career plan

Level 3: Strategic Planner (Strategic)

Achievement goal: Can strategically plan real-world MI applications and take concrete actions toward career goals

Level 4: Leader/Expert (Leadership)

Achievement goal: - Recognized as MI thought leader - Can influence multiple stakeholders - Can nurture next generation of MI researchers and engineers

6.9 Readiness Check for Next Steps

Preparation for Practical Projects

Preparation for Academic Research

Preparation for Entrepreneurship/Startups

Preparation for Global Expansion

6.10 Action Plan for Self-Growth

Execute This Week

Execute This Month

Execute Within 3 Months

Execute Within 1 Year

Tips for using checklist:

1. Review regularly — Check progress monthly against career plan.

2. Prioritize unachieved items — Improve overall ability by overcoming weaknesses.

3. Record level determination — Aim for level up every 3 months.

4. Mentor/peer review — Seek feedback from others.

5. Use in practice — Use for self-evaluation during job hunting and grant applications.

Exercises

Problem 1 (Difficulty: easy)

Select one of the five cases introduced in this chapter and explain the following: - What challenges existed - How was MI utilized - What results were obtained

Hint Consider Case Study 2 (catalyst materials) as an example. There was a clear challenge of finding alternative materials to platinum.
Answer Example (Catalyst Materials Case) **Challenge**: Catalysts used in hydrogen production and fuel cells require platinum (Pt), which is expensive (approximately 4,000 yen/g) and rare, necessitating low-cost, high-activity alternative catalysts. **MI Utilization**: - Predict hydrogen adsorption energy using first-principles calculation (DFT) - Efficiently select next experimental candidates using Bayesian Optimization - Discover optimal composition in 10-20 experiments **Results**: - Mo-Co-N system: reduced Pt usage by 50%, activity 120% - Development period: 2 years → 3 months (approximately 8x faster) - Cost reduction: catalyst price reduced by 80% (1 million yen/kg → 200,000 yen/kg)

Problem 2 (Difficulty: medium)

Compare self-driving labs with conventional human-led laboratories and list three advantages and disadvantages for each.

Hint Consider from perspectives of speed, cost, and creativity.
Answer Example **Self-Driving Lab Advantages**: 1. **24-hour operation**: no human labor hour constraints, experiments continue on holidays 2. **Acceleration**: synthesize and evaluate 41 materials in 17 days (approximately 10x human speed) 3. **Reproducibility**: minimize experimental errors through precise robot control **Self-Driving Lab Disadvantages**: 1. **High initial investment**: approximately 100 million yen equipment introduction cost 2. **Low flexibility**: difficult to automate complex synthesis procedures (high-temperature processing, etc.) 3. **Lack of creativity**: difficult for human intuitive discoveries **Conventional Laboratory Advantages**: 1. **Flexibility**: can respond immediately to unexpected results 2. **Creativity**: can try new ideas with human intuition 3. **Low initial cost**: utilize existing equipment and personnel **Conventional Laboratory Disadvantages**: 1. **Labor hour constraints**: operate only 8 hours/day, 5 days/week 2. **Reproducibility issues**: errors easily occur due to experimenter variation 3. **Low throughput**: can evaluate only about 10-100 materials per year

Problem 3 (Difficulty: medium)

In a materials field of interest (batteries, catalysts, semiconductors, polymers, etc.), propose how MI can be utilized with a concrete project plan. Include the following: - Problem definition - MI approach (methods to use) - Expected outcomes

Hint Apply examples from this chapter to your field of interest.
Answer Example (Semiconductor Materials Case) **Field**: Transparent Conductive Oxide (TCO) **Challenge**: - Smartphone touch panels require transparent and highly conductive materials - Current mainstream material ITO (Indium Tin Oxide) uses rare and expensive indium - Difficult to balance transparency (visible light transmittance >80%) and conductivity (resistivity <10^-4 Ω·cm) **MI Approach**: 1. **Data collection**: Obtain bandgap and electrical conductivity data for 100,000 oxide materials from Materials Project 2. **Prediction model construction**: Predict the following using Graph Neural Networks (GNN) - Bandgap (transparency indicator: 3.0-3.5 eV optimal) - Carrier concentration (conductivity indicator) 3. **Screening**: Narrow from 100,000 types to 100 types that balance transparency and conductivity 4. **Multi-objective optimization**: Also consider cost (avoid rare elements) 5. **Experimental validation**: Synthesize and measure top 10 types **Expected Outcomes**: - Discovery of indium-free TCO (e.g., Sn-Zn-O system) - 50% material cost reduction - Development period: 5 years → 1 year (80% reduction) - Contribution to touch panel market (annual market size approximately 5 trillion yen)

Problem 4 (Difficulty: hard)

If you were to choose "academia," "industry," or "startup" career paths, which would you choose? Explain your reasons from the following perspectives: - Salary and economic rewards - Research freedom - Social impact - Lifestyle - Personal values

Hint There is no correct answer. Organize your values.
Answer Example (Choosing Industry) **Choice**: Industry (MI Engineer at major chemical manufacturer) **Reasons**: **1. Salary and Economic Rewards**: - Higher salary than academia (annual 7-10M yen vs. 5-7M yen) - Stable employment (for large companies) - Economic stability important for supporting family **2. Research Freedom**: - Themes align with company business strategy, but MI field is broad enough to be acceptable - Short-term results required, but that becomes my motivation **3. Social Impact**: - Direct impact on society as products reach market - Example: battery material improvement → EV adoption → CO2 reduction - More attractive "visible form" contribution than academic papers **4. Lifestyle**: - Want to avoid academia's long working hours (nighttime/weekend research) - Emphasize work-life balance (time with family) - Industry (varies by company) but relatively stable schedule **5. Personal Values**: - More interested in "solving social issues" than "research for research sake" - Value team achievements over academic competition (paper count, citations) - Want sense of accomplishment in 10 years: "products I worked on are used worldwide" **Conclusion**: Want to contribute to practical materials development while maintaining stable life as industry MI Engineer. However, keep option of future startup transition, continuously learning latest technologies.

Problem 5 (Difficulty: hard)

We stated that "sustainability-driven design" will become important as a future MI trend. Design an MI project considering sustainability in a materials field of interest. Include the following: - Specific environmental burden indicators (CO2 emissions, toxicity, recyclability, etc.) - How to handle performance and sustainability tradeoffs - Social and economic impact

Hint Apply Section 3.3 "Sustainability-Driven Design" to a specific material system.
Answer Example (Plastic Packaging Materials Case) **Project Name**: Multi-objective Optimization of Biodegradable Plastics **Challenge**: - Global plastic waste: 300 million tons/year, ocean leakage 10 million tons - Conventional plastics (PE, PP) take hundreds of years to decompose - Biodegradable plastics (PLA, PHA) have low performance (strength, heat resistance) **Environmental Burden Indicators**: 1. **CO2 emissions**: Carbon footprint during manufacturing (kg-CO2/kg) - Conventional PE: approximately 2.0 kg-CO2/kg - Target: < 1.0 kg-CO2/kg 2. **Biodegradability**: Decomposition rate after 6 months (%) - Conventional PE: < 5% - Target: > 90% 3. **Toxicity**: Toxicity to microorganisms and aquatic life (LC50 value) - Conventional PE: low toxicity but microplastics problematic - Target: completely harmless (including decomposition products) **Performance Indicators**: - Tensile strength: > 30 MPa (PE is 35 MPa) - Heat resistance: > 80°C (food packaging use) - Cost: < 300 yen/kg (PE is 200 yen/kg) **MI Approach**: 1. **Data collection**: - Polymer literature data (5,000 types) - Life Cycle Assessment (LCA) database 2. **Multi-objective optimization model**: - Predict strength, heat resistance, biodegradability using Random Forest - Visualize Pareto front (performance vs. environmental burden tradeoff) 3. **Constraints**: - Exclude toxic substances (phthalates, BPA, etc.) - Don't use rare elements 4. **Experimental validation**: - Select 10 types from Pareto optimal solutions - Synthesis, measurement, LCA evaluation **Tradeoff Handling**: - **Case 1 (High performance focus)**: Strength 35 MPa, decomposition rate 70%, CO2 1.2 kg-CO2/kg - Use: Industrial packaging (recycle after short-term use) - **Case 2 (Environmental focus)**: Strength 28 MPa, decomposition rate 95%, CO2 0.8 kg-CO2/kg - Use: Agricultural mulch film (decomposes in soil) - **Case 3 (Balance type)**: Strength 32 MPa, decomposition rate 85%, CO2 1.0 kg-CO2/kg - Use: Food packaging (convenience store lunch boxes, etc.) **Expected Outcomes**: - 50% CO2 emission reduction while maintaining performance - Mitigate ocean plastic problem - Market size: biodegradable plastic market predicted to reach 1 trillion yen in 2030 - Regulatory compliance: meets EU plastic regulations **Social and Economic Impact**: - Environmental: protect marine ecosystems, reduce carbon emissions - Economic: create new markets, generate employment - Policy: contribute to SDG Goal 12 (sustainable consumption and production), Goal 14 (ocean resources)

References

Success Stories

  1. Chen, C., Zuo, Y., Ye, W., Li, X., Deng, Z., & Ong, S. P. (2020). "A critical review of machine learning of energy materials." Advanced Energy Materials, 10(8), 1903242. DOI: 10.1002/aenm.201903242

  2. Nørskov, J. K., Bligaard, T., Rossmeisl, J., & Christensen, C. H. (2009). "Towards the computational design of solid catalysts." Nature Chemistry, 1(1), 37-46. DOI: 10.1038/nchem.121

  3. Huang, W., Martin, P., & Zhuang, H. L. (2019). "Machine-learning phase prediction of high-entropy alloys." Acta Materialia, 169, 225-236. DOI: 10.1016/j.actamat.2019.03.012

  4. Mannodi-Kanakkithodi, A., Chandrasekaran, A., Kim, C., Huan, T. D., Pilania, G., Botu, V., & Ramprasad, R. (2018). "Scoping the polymer genome: A roadmap for rational polymer dielectrics design and beyond." Materials Today, 21(7), 785-796. DOI: 10.1016/j.mattod.2017.11.021

  5. Agrawal, A., & Choudhary, A. (2016). "Perspective: Materials informatics and big data: Realization of the fourth paradigm of science in materials science." APL Materials, 4(5), 053208. DOI: 10.1063/1.4946894

Future Trends

  1. Szymanski, N. J., Rendy, B., Fei, Y., et al. (2023). "An autonomous laboratory for the accelerated synthesis of novel materials." Nature, 624(7990), 86-91. DOI: 10.1038/s41586-023-06734-w

  2. Chen, C., & Ong, S. P. (2022). "A universal graph deep learning interatomic potential for the periodic table." Nature Computational Science, 2(11), 718-728. DOI: 10.1038/s43588-022-00349-3

  3. Olivetti, E. A., Cole, J. M., Kim, E., Kononova, O., Ceder, G., Han, T. Y. J., & Hiszpanski, A. M. (2020). "Data-driven materials research enabled by natural language processing and information extraction." Applied Physics Reviews, 7(4), 041317. DOI: 10.1063/5.0021106

Career and Education

  1. Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O., & Walsh, A. (2018). "Machine learning for molecular and materials science." Nature, 559(7715), 547-555. DOI: 10.1038/s41586-018-0337-2

  2. Ramprasad, R., Batra, R., Pilania, G., Mannodi-Kanakkithodi, A., & Kim, C. (2017). "Machine learning in materials informatics: recent applications and prospects." npj Computational Materials, 3(1), 54. DOI: 10.1038/s41524-017-0056-5

Online Resources

  1. Materials Project: https://materialsproject.org
  2. Citrine Informatics: https://citrine.io
  3. Kebotix: https://www.kebotix.com
  4. Matmerize: https://www.matmerize.com
  5. MRS (Materials Research Society): https://www.mrs.org

Author Information

This article was created as part of the MI Knowledge Hub project under the supervision of Dr. Yusuke Hashimoto, Tohoku University.

Series Information:

Update History:

← Previous Chapter Return to Series Index

Disclaimer