C9259 - AI AND DATA DRIVEN MATERIALS

Learning outcomes

This course explores the synergy between machine learning, artificial intelligence, and materials research, aiming to accelerate the discovery, optimization, and deployment of advanced materials and technologies. Students will explore data-driven methodologies to design and analyze materials and solutions for next-generation batteries, with a primary focus on Lithium-ion systems. Key topics include data collection and mining, machine learning techniques, data-driven materials discovery and structure-property relationships. The course includes practical, hands-on labs where students will apply ML techniques to selected problems and case studies. By the end of the course, participants will be equipped with the skills to harness AI and ML tools to drive innovation in battery research and development.

Course contents

Philosophy of the course

The course introduces modern artificial intelligence and data-driven approaches for materials science through the lens of the scientific method. Rather than presenting machine-learning algorithms as isolated techniques, the course is organised around the key scientific questions encountered during a materials-discovery workflow: how materials are transformed into data, how predictive models are constructed and validated, how new materials can be discovered, and how artificial intelligence can support scientific understanding. Throughout the course, emphasis is placed on physical interpretation, reproducibility, uncertainty quantification, and responsible use of AI.

Artificial intelligence does not replace scientific reasoning; it augments it. The ultimate goal is not simply to predict materials properties, but to generate new scientific understanding.

 

Module 1: From Materials to Data

How can materials be transformed into scientific data?

Topics:

• Materials discovery in the age of AI

• The materials discovery cycle

• Experimental, computational and literature data

• Materials databases and FAIR data

• Data quality, metadata and reproducibility

• Applications to energy materials and batteries

Laboratory

Exploration and analysis of a real materials dataset.

 

Module 2: From Data to Materials Representation

How can a material be represented so that AI can learn from it?

Topics:

• Composition-based representations

• Crystal-structure representations

• Physically motivated descriptors

• Structural fingerprints

• Feature engineering

• Graph-based representations

• Advantages and limitations of different descriptors

Laboratory

Construction and comparison of different materials representations.

 

Module 3: From Representation to Prediction

How can predictive models be built?

Topics:

• Regression and classification

• Machine-learning workflow

• Training, validation and testing

• Model selection

• Performance metrics

• Classical and modern machine-learning algorithms

• Neural networks and graph neural networks

Laboratory

Prediction of materials properties using different machine-learning models.

 

Module 4: From Prediction to Scientific Understanding

When can an AI model be trusted?

Topics:

• Cross-validation
• Hyperparameter optimisation
• Model interpretability
• Feature importance
• Uncertainty quantification
• Domain of applicability
• Physical consistency
• Explainable AI

Laboratory

Validation, uncertainty analysis and interpretation of machine-learning models.

 

Module 5: From Prediction to Discovery

How can AI accelerate the discovery of new materials?

Topics:

• High-throughput materials screening

• Surrogate models

• Multi-objective optimisation

• Active learning

• Bayesian optimisation

• Inverse materials design

Laboratory

Screening and ranking of candidate materials.

 

Module 6: AI for Energy Materials and Batteries

How can AI contribute to next-generation energy materials?

Topics:

• Data-driven battery materials research

• Electrode and electrolyte materials

• Prediction of voltage, stability, ionic conductivity and capacity

• Degradation and lifetime prediction

• Battery informatics

• Representative case studies

Laboratory

Machine-learning workflow applied to battery materials.

 

Module 7: AI for Scientific Discovery

How should scientists use Generative AI?

Topics:

• Generative AI and Large Language Models
• Literature exploration
• Scientific coding
• Workflow generation
• Retrieval-augmented generation
• Verification of AI-generated results
• Hallucinations and scientific reliability
• Responsible use of AI in research

Laboratory

Integrated AI-assisted materials-informatics project.

 

Readings/Bibliography

Lecture notes

Books on AI/ML Materials Informatics

Materials Data Science, Stefan Sandfeld, Springer (2024)

Machine Learning for Materials Discovery, N. M. Anoop Krishnan, Hariprasad Kodamana, Ravinder Bhattoo, Springer (2024)

Specific on ML:

Data-driven modeling & scientific computation: methods for complex systems & big data, J. N. Kutz, Oxford University Press (2013)

Hands-on Machine Learning with Scikit-Learn, Keras & TensorFlow, Aurelien Geron, O'Reilly (2019)

Teaching methods

Lecture (Blackboard & projector), computer lab, problem solving, Group work.

Given the nature of the activities and the teaching methods adopted, attendance of this course requires all students to have previously completed Modules 1 and 2 of the e-learning training on health and safety in study environments.

Assessment methods

1. Evaluation of the Lab reports (50%)

During the course, students will participate in compulsory guided laboratory activities, at the end of which each student will be required to prepare a report, either individually or as part of a group, following the instructions and any templates provided by the instructor. The report must describe the activities carried out, present the results obtained, and include a critical analysis of them.

Reports must normally be submitted before the beginning of the examination period.

The reports will be assessed according to the following criteria:

• completeness and accuracy in carrying out the assigned tasks;
• correctness of the data analysis and interpretation;
• ability to present the results clearly and in a well-structured manner;
• proficiency in the use of the tools and methodologies employed.
• ability to verify and critically evaluate AI-generated results.

The score obtained for this component will account for 50% of the final grade.

2. Oral examination (50%)

The oral examination consists of a discussion and defense of the laboratory reports, together with questions on the theoretical concepts covered during the lectures. The assessment focuses on the student’s ability to:

• explain the physical, computational, and methodological assumptions;
• present and discuss the results obtained;
• assess the reliability and physical consistency of AI-generated results;
• identify limitations, methodological weaknesses, and possible sources of error;
• modify, correct, or debug part of the workflow during the examination;
• connect the practical activities with the theoretical content of the course.

 

The oral examination may be taken only after the laboratory reports have been submitted and positively assessed, as successful completion of the theoretical-practical component is a prerequisite for taking the theoretical component.

The score obtained in the oral examination accounts for 50% of the final grade.

 

AI statement

As regards the assessment of learning outcomes, the examination consists of one component involving substantial use of AI, such as problem solving and content generation, and a compulsory critical analysis component.

 

Assessment Criteria and Final Grade

The final grade, expressed on a 30-point scale, is calculated as the weighted average of the marks obtained for the laboratory reports and the oral examination, each contributing 50%.

Honours may be awarded when an excellent level of achievement is demonstrated in both components.

The final grade is assigned according to the following general criteria:

• Below 18: insufficient knowledge and/or skills; failure to achieve the minimum learning objectives;
• 18–21: achievement of the minimum required knowledge and skills; basic presentation;
• 22–24: satisfactory competence; clear and adequately reasoned presentation;
• 25–27: good command of the subject matter, autonomy, and ability to apply the acquired knowledge;
• 28–30: thorough command of the subject matter, clear and well-structured presentation, and excellent critical skills;
• 30 with honours: outstanding achievement, a high degree of autonomy, strong ability to pursue further independent analysis, and exceptional critical skills.

Students with specific learning disabilities (SLDs), ADHD, or other temporary or permanent disabilities are advised to contact the relevant University office well in advance. The office will propose any appropriate accommodations for the students concerned. These accommodations must, in any case, be submitted to the instructor for approval at least 15 days in advance. The instructor will assess their suitability, also in relation to the learning objectives of the course.

Teaching tools

Teaching and computing materials provided by the instructor.

Office hours

See the website of Cesare Franchini