50 lines (39 loc) · 1.71 KB

Week 10 Summary: ML for characterization signals

Cross-Book Summary

1. Clustering Spectral Data

K-Means: Groups similar spectra (XRD/EDS) to identify distinct phases.
Mini-Batch K-Means: Speeds up high-throughput characterization.
t-SNE: Projects high-dimensional spectra to 2D to reveal outliers/relationships.

2. Autoencoders for Signal Processing

Latent Representations: Compresses spectra to essential physical information.
Denoising: Reconstructs clean signals from noisy inputs without blurring.
Non-linear Compression: Outperforms PCA for complex spectral libraries.

3. Scientific Integrity in ML

Peak Preservation: ML must assist, not invent or smooth away real physics.

90-Minute Lecture Strategy

Part 1: High-Dimensional Signals

Digital footprint: XRD, EDS, EELS, Raman.
Manual vs. automated peak-picking.
Vector spectrum representation.

Part 2: Clustering Structure

K-Means algorithm.
Elbow Method for phase counting.
Ternary alloy mapping.

Part 3: Visualizing the Unseen

t-SNE Stochastic Proximity.
Hidden relationships.
t-SNE distance pitfalls.

Part 4: Autoencoders & Denoising

Encoder-Bottleneck-Decoder.
Denoising characterization signals.
Bottlenecks as physical descriptors.

Part 5: Data to Discovery

Real-time spectral analysis.
Physical consistency in ML.
Automated pipelines.

Quarto Website Update (Summary)

Summary for ML-PC Week 10:

Processes high-dimensional Characterization Signals (XRD, EDS).
Employs K-Means and t-SNE for automated phase identification.
Uses Autoencoders for latent space compression and denoising.
Enhances high-throughput data analysis while preserving physics.