Ai4MatLectures

PyTorch teaching notebooks for materials science

Teaching notebooks for the ECLIPSE Lab lecture triad at FAU Erlangen-Nürnberg. Install the package:

pip install git+https://github.com/ECLIPSE-Lab/Ai4MatLectures.git "mdsdata>=0.1.5"

Live (braided) weekly notebooks

These are the canonical in-class notebooks — one homework + one main file per week, braiding the matching MFML and ML-PC units. Run them as Python scripts (Jupytext cell markers).

Week	Topic	Homework	Main (90 min)
3	Loss minimization & leakage	week3_homework.py	week3_loss_and_leakage.py
4	Architecture & representation	week4_homework.py	week4_architecture_and_representation.py
5	Clustering & autoencoders	week5_homework.py	week5_clustering_and_autoencoders.py
6	Optimization & fine-tuning	week6_homework.py	week6_optimization_and_finetuning.py
7	Generalization, ensembles & distribution shift	week7_homework.py	week7_generalization_and_ensembles.py
8	Uncertainty, robustness & graph-based crystal reps	week8_uncertainty_and_robustness_homework.py	week8_uncertainty_and_robustness.py
9	Latent geometry — PCA, t-SNE, probing, contrastive	week9_homework.py	week9_latent_geometry.py
10	Attention, ViT & cross-system transfer	week10_homework.py	week10_attention_and_transfer.py
11	Generative inverse design (VAE, CVAE, DDPM)	week11_homework.py	week11_generative_inverse_design.py
12	Uncertainty, GPs & discovery	week12_homework.py	week12_uncertainty_and_discovery.py
13	Physics-informed learning, GPs & active discovery	week13_homework.py	week13_physics_and_gp.py
14	Explainability, limits & course retrospective	week14_homework.py	week14_explainability_and_limits.py

MFML — Mathematical Foundations of AI & ML

Week	Topic	Dataset	Notebook
4	NN architecture & backprop	Iris / Alpaydin Digits	week04_classifier_iris, week05_backprop_digits
5	Clustering & autoencoders	Nanoindentation / Ising full	braided: week5_clustering_and_autoencoders.py; per-course: week11_clustering_nanoindentation, week10_autoencoder_ising_full
7	Generalization, regularization & tree ensembles	Ising light / Tensile / SOAP	braided: week7_generalization_and_ensembles.py; per-course: week07_overfitting_ising_light
8	Probabilistic view, MLE/MAP, calibration	Tensile / 1-D toy	braided: week8_uncertainty_and_robustness.py (Blocks 2–5)
10	Latent spaces & embeddings	Ising full	week10_autoencoder_ising_full
11	Unsupervised learning (deep dive)	Nanoindentation	week11_clustering_nanoindentation
12	Uncertainty in predictions (GPs)	Tensile / 1-D toy	braided: week12_uncertainty_and_discovery.py

MLPC — ML in Materials Processing & Characterization

Week	Topic	Dataset	Notebook
4	Microstructure representations & CNN baseline	Alpaydin Digits / Ising	week04_baseline_digits, week05_cnn_ising_light, week05_cnn_ising_full
5	Unsupervised learning in materials	Ising / Cahn-Hilliard / NEU-DET / ESTM	braided: week5_clustering_and_autoencoders.py; per-course: week05_clustering_estm, week05_clustering_neu_det, week11_anomaly_cahn_hilliard
7	Time-series / process monitoring (lecture cancelled — Pfingstdienstag)	Tensile Test	supplementary: week07_process_monitoring_tensile; the cross-T generalisation block in week7_generalization_and_ensembles.py also previews the Week 8 lecture
8	Generalization, robustness & process windows (exercise cancelled — Fronleichnam)	Tensile Test	braided self-study: week7_generalization_and_ensembles.py (Blocks 2–5) and week8_uncertainty_and_robustness.py (Blocks 6–7, sensitivity + process windows)
11	Anomaly detection via AE (automation context)	Cahn-Hilliard	week11_anomaly_cahn_hilliard
12	UQ + Bayesian active learning	Tensile Test	braided: week12_uncertainty_and_discovery.py (Blocks 2–4)
13	MC-Dropout segmentation UQ (self-study — lecture cancelled 07.07)	MetalDAM	week13_mcdropout_metaldam

MG — Materials Genomics

Week	Topic	Dataset	Notebook
5	Descriptors + regression	Chemical Elements	week05_descriptors_elements
6	Local atomic envs (SOAP) & universal MLIPs (MACE-MP-0) — Unit 6 after the u06↔︎u07 swap	ASE bulk prototypes (Cu/Fe/Al/Si/NaCl/MgO)	braided: week6_optimization_and_finetuning.py (Block 6 — frozen-SOAP regression + MLIP-as-pretrained-backbone); per-course standalone: week06_soap_and_mace (SOAP + MACE-MP-0 EOS benchmark + MLIP-MD)
7	Lecture cancelled (Pfingstdienstag) — descriptor recap now lives in Week 6	—	—
8	Graph-based crystal representations (Unit 7) — Week 8 after the u06↔︎u07 swap	CrystalGraphs (toy) + PBC graphs	braided anchor: week8_uncertainty_and_robustness.py (Blocks 8/8b — TinyCGNN, PBC neighbour search, RBF + cutoff artifact, ranking metrics)
9	Regression & generalization in materials data	Nanoindentation	week09_regression_nanoindentation
11	Latent space (Ising)	Ising light	week11_latent_ising
11	Materials latent space	Cahn-Hilliard	week11_latent_cahn_hilliard
12	Clustering vs discovery + per-cluster GPs	Nanoindentation	braided: week12_uncertainty_and_discovery.py (Blocks 5–6)