ML-PC: cascade week summaries +1 after Week-7 cancellation (week7 cancelled; buffer dropped; week14 = reflection)

Author: PhilippPelz

week10_summary.md

@@ -1,49 +1,50 @@
-# Week 10 Summary: Transformers for materials characterization
+# Week 10 Summary: ML for characterization signals
 
 ## Cross-Book Summary
 
-### 1. Why Attention for Materials
-- **Long-range correlations:** Many characterization signals (diffraction, micrograph stacks) have dependencies that exceed practical CNN receptive fields.
-- **Self-attention:** Lets every token attend to every other token directly, capturing global structure in one layer.
+### 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. The Transformer Toolkit
-- **Scaled dot-product attention:** The core operation and its O(L²) cost.
-- **Vision Transformer (ViT):** Patchify → embed → encode → classify; transformers applied to image-like data.
-- **Flash Attention:** A fused kernel that makes long sequences tractable without materialising the L×L matrix.
+### 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. Materials Applications
-- **ViT on 4D-STEM:** Diffraction patches become a token sequence for a ViT encoder.
-- **Cross-attention across LPBF layers:** Long-stack micrograph context for additive-manufacturing monitoring.
-
-### 4. Scaling Alternatives (Awareness Only)
-- **Mamba / structured state-space models (SSMs):** O(L) compute, constant memory; competitive on long sequences. Cross-reference the Week 7 time-series deck.
+### 3. Scientific Integrity in ML
+- **Peak Preservation:** ML must assist, not invent or smooth away real physics.
 
 ## 90-Minute Lecture Strategy
 
-### Part 1: Where We Are
-- Recap of Week 9 (characterization signals) and why we now need attention.
+### Part 1: High-Dimensional Signals
+- Digital footprint: XRD, EDS, EELS, Raman.
+- Manual vs. automated peak-picking.
+- Vector spectrum representation.
 
-### Part 2: Why Attention
-- Long-range correlations exceed CNN receptive fields.
+### Part 2: Clustering Structure
+- K-Means algorithm.
+- Elbow Method for phase counting.
+- Ternary alloy mapping.
 
-### Part 3: Mechanics
-- Scaled dot-product attention: the formula and the cost.
-- ViT in five lines: patchify, embed, encode, classify.
-- Flash Attention: the kernel that makes long sequences tractable.
+### Part 3: Visualizing the Unseen
+- t-SNE Stochastic Proximity.
+- Hidden relationships.
+- t-SNE distance pitfalls.
 
-### Part 4: Materials Applications
-- ViT on 4D-STEM diffraction.
-- Cross-attention across LPBF layer stacks.
+### Part 4: Autoencoders & Denoising
+- Encoder-Bottleneck-Decoder.
+- Denoising characterization signals.
+- Bottlenecks as physical descriptors.
 
-### Part 5: Practice and Pitfalls
-- `nn.MultiheadAttention` vs `F.scaled_dot_product_attention`.
-- Scaling alternatives (Mamba / SSMs) — mention only.
-- Anti-patterns: what *not* to do.
-- Exercise preview.
+### 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:**
-- Motivates self-attention for long-range structure in characterization data.
-- Covers scaled dot-product attention, the Vision Transformer, and Flash Attention.
-- Applies transformers to 4D-STEM diffraction and LPBF layer-stack context.
-- Notes Mamba / state-space models as scaling alternatives, and when not to reach for a transformer.
+**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.

week11_summary.md

@@ -1,47 +1,49 @@
-# Week 11 Summary: Uncertainty-aware regression & Gaussian Processes
+# Week 11 Summary: Transformers for materials characterization
 
 ## Cross-Book Summary
 
-### 1. Knowing what you don't know
-- **Aleatoric vs. Epistemic:** Inherent physical noise vs. model ignorance.
-- **Overconfidence Danger:** Point estimates fail safely in unknown regimes; uncertainty metrics are crucial.
+### 1. Why Attention for Materials
+- **Long-range correlations:** Many characterization signals (diffraction, micrograph stacks) have dependencies that exceed practical CNN receptive fields.
+- **Self-attention:** Lets every token attend to every other token directly, capturing global structure in one layer.
 
-### 2. Gaussian Processes (GPs)
-- **Distribution over Functions:** GP yields posterior mean and variance (uncertainty).
-- **Kernels as Physical Priors:** Encodes assumptions about data smoothness/scale.
-- **Non-Parametric Nature:** Scales with data size, ideal for small, high-quality materials datasets.
+### 2. The Transformer Toolkit
+- **Scaled dot-product attention:** The core operation and its O(L²) cost.
+- **Vision Transformer (ViT):** Patchify → embed → encode → classify; transformers applied to image-like data.
+- **Flash Attention:** A fused kernel that makes long sequences tractable without materialising the L×L matrix.
 
-### 3. GP-Based Process Maps
-- **Confidence Ribbons:** Visualize reliability to guide further experiments.
-- **Kriging:** Interpolates materials property surfaces using GP regression.
+### 3. Materials Applications
+- **ViT on 4D-STEM:** Diffraction patches become a token sequence for a ViT encoder.
+- **Cross-attention across LPBF layers:** Long-stack micrograph context for additive-manufacturing monitoring.
+
+### 4. Scaling Alternatives (Awareness Only)
+- **Mamba / structured state-space models (SSMs):** O(L) compute, constant memory; competitive on long sequences. Cross-reference the Week 8 time-series deck.
 
 ## 90-Minute Lecture Strategy
 
-### Part 1: Uncertainty in Science
-- Risk management in materials processing.
-- Visualizing distributions and error bars.
+### Part 1: Where We Are
+- Recap of Week 10 (characterization signals) and why we now need attention.
 
-### Part 2: GP Fundamentals
-- Function vs. Parameter space.
-- Kernels and "Similarity".
-- Conditional Gaussians and Variance.
+### Part 2: Why Attention
+- Long-range correlations exceed CNN receptive fields.
 
-### Part 3: GP Case Studies
-- Predicting tensile strength across parameters.
-- GP for Experimental Design.
-- Multi-Task GPs.
+### Part 3: Mechanics
+- Scaled dot-product attention: the formula and the cost.
+- ViT in five lines: patchify, embed, encode, classify.
+- Flash Attention: the kernel that makes long sequences tractable.
 
-### Part 4: Advanced Probabilistic ML
-- Mixture Density Networks (MDNs).
-- Dropout as Bayesian approximation.
+### Part 4: Materials Applications
+- ViT on 4D-STEM diffraction.
+- Cross-attention across LPBF layer stacks.
 
-### Part 5: Decision Making
-- Safe process windows via confidence intervals.
-- Building trustworthy models.
+### Part 5: Practice and Pitfalls
+- `nn.MultiheadAttention` vs `F.scaled_dot_product_attention`.
+- Scaling alternatives (Mamba / SSMs) — mention only.
+- Anti-patterns: what *not* to do.
+- Exercise preview.
 
 ## Quarto Website Update (Summary)
-**Summary for ML-PC Week 12:**  
-- Introduces Probabilistic Machine Learning for uncertainty quantification.
-- Differentiates aleatoric (noise) from epistemic (ignorance) uncertainty.
-- Uses Gaussian Processes (GPs) for uncertainty-aware regression.
-- Applies confidence intervals to map robust process windows.
+**Summary for ML-PC Week 11:**
+- Motivates self-attention for long-range structure in characterization data.
+- Covers scaled dot-product attention, the Vision Transformer, and Flash Attention.
+- Applies transformers to 4D-STEM diffraction and LPBF layer-stack context.
+- Notes Mamba / state-space models as scaling alternatives, and when not to reach for a transformer.

week12_summary.md

@@ -1,49 +1,47 @@
-# Week 12 Summary: Physics-informed and constrained ML
+# Week 12 Summary: Uncertainty-aware regression & Gaussian Processes
 
 ## Cross-Book Summary
 
-### 1. Physics-Informed Neural Networks (PINNs)
-- **Embedding Laws:** Enforce ODEs/PDEs via the loss function for physical consistency.
-- **Automatic Differentiation:** Exact derivative calculations enable NNs to evaluate physical equations.
-- **Boundary Conditions:** Methods like Lagaris substitution guarantee boundary compliance.
+### 1. Knowing what you don't know
+- **Aleatoric vs. Epistemic:** Inherent physical noise vs. model ignorance.
+- **Overconfidence Danger:** Point estimates fail safely in unknown regimes; uncertainty metrics are crucial.
 
-### 2. Governing Equation Discovery
-- **Dictionary-Based Regression:** Build a dictionary of candidate math functions.
-- **Sparse Identification:** Use regularized regression (Lasso) to discover physical laws from noisy data.
-- **Dimensional Reasoning:** Unit analysis ensures physically plausible discoveries.
+### 2. Gaussian Processes (GPs)
+- **Distribution over Functions:** GP yields posterior mean and variance (uncertainty).
+- **Kernels as Physical Priors:** Encodes assumptions about data smoothness/scale.
+- **Non-Parametric Nature:** Scales with data size, ideal for small, high-quality materials datasets.
 
-### 3. Constraints in Materials Science
-- **Monotonicity:** Enforce required physical trends (e.g., hardness vs. alloying).
-- **Hybrid Modeling:** Combine physical "White-Box" models with data-driven "Black-Boxes" (Grey-Box).
+### 3. GP-Based Process Maps
+- **Confidence Ribbons:** Visualize reliability to guide further experiments.
+- **Kriging:** Interpolates materials property surfaces using GP regression.
 
 ## 90-Minute Lecture Strategy
 
-### Part 1: Why Physics Matters
-- Limits of unconstrained Black-Box models.
-- Accurate but Physical models.
-- PINNs need less data.
+### Part 1: Uncertainty in Science
+- Risk management in materials processing.
+- Visualizing distributions and error bars.
 
-### Part 2: Automatic Differentiation
-- GradientTape mechanics.
-- Derivatives as ML architecture components.
+### Part 2: GP Fundamentals
+- Function vs. Parameter space.
+- Kernels and "Similarity".
+- Conditional Gaussians and Variance.
 
-### Part 3: Solving Physics with NNs
-- PINN Architectures: Data Loss + Physics Loss.
-- Enforcing Boundary Conditions.
-- 3D printing heat transfer case study.
+### Part 3: GP Case Studies
+- Predicting tensile strength across parameters.
+- GP for Experimental Design.
+- Multi-Task GPs.
 
-### Part 4: Equation Discovery
-- Sparse Regression and candidate dictionaries.
-- Damped pendulum equation case study.
-- Unit Analysis search pruning.
+### Part 4: Advanced Probabilistic ML
+- Mixture Density Networks (MDNs).
+- Dropout as Bayesian approximation.
 
-### Part 5: The Grey-Box Future
-- Hybrid architectures vs. FEA.
-- Building industrial trust.
+### Part 5: Decision Making
+- Safe process windows via confidence intervals.
+- Building trustworthy models.
 
 ## Quarto Website Update (Summary)
-**Summary for ML-PC Week 13:**  
-- Combines neural networks with physical laws via Physics-Informed ML.
-- Introduces PINNs and automatic differentiation.
-- Details Governing Equation Discovery using sparse regression.
-- Applies physical constraints to build data-efficient Grey-Box models.
+**Summary for ML-PC Week 12:**  
+- Introduces Probabilistic Machine Learning for uncertainty quantification.
+- Differentiates aleatoric (noise) from epistemic (ignorance) uncertainty.
+- Uses Gaussian Processes (GPs) for uncertainty-aware regression.
+- Applies confidence intervals to map robust process windows.

week13_summary.md

@@ -1,49 +1,49 @@
-# Week 13 Summary: Integration, limits, and reflection
+# Week 13 Summary: Physics-informed and constrained ML
 
 ## Cross-Book Summary
 
-### 1. Explainability: Opening the Black Box
-- **Beyond Prediction:** "Why" builds industrial trust over just "what".
-- **Sensitivity Analysis:** Perturb inputs to find primary process drivers.
-- **Levels of Explanation:** Tailor explanations for managers, experts, and data scientists.
+### 1. Physics-Informed Neural Networks (PINNs)
+- **Embedding Laws:** Enforce ODEs/PDEs via the loss function for physical consistency.
+- **Automatic Differentiation:** Exact derivative calculations enable NNs to evaluate physical equations.
+- **Boundary Conditions:** Methods like Lagaris substitution guarantee boundary compliance.
 
-### 2. Causality and Semantics
-- **Causal Process Chains:** Shift from anomaly detection to early prediction.
-- **Ontologies:** Digitize semantic meaning (e.g., mapping variables to physical concepts) to enable algorithmic reasoning.
+### 2. Governing Equation Discovery
+- **Dictionary-Based Regression:** Build a dictionary of candidate math functions.
+- **Sparse Identification:** Use regularized regression (Lasso) to discover physical laws from noisy data.
+- **Dimensional Reasoning:** Unit analysis ensures physically plausible discoveries.
 
-### 3. Limits of AI in Materials Science
-- **Data Bias:** Models trained only on "successes" cannot predict failures.
-- **AI Hallucinations:** Large models may generate physically impossible patterns.
-- **Expert's Role:** AI automates analysis, but human experts define questions and interpret final truths.
+### 3. Constraints in Materials Science
+- **Monotonicity:** Enforce required physical trends (e.g., hardness vs. alloying).
+- **Hybrid Modeling:** Combine physical "White-Box" models with data-driven "Black-Boxes" (Grey-Box).
 
 ## 90-Minute Lecture Strategy
 
-### Part 1: Course Synthesis
-- Recap: Signal formation to PINNs.
-- AI-driven materials lifecycle.
+### Part 1: Why Physics Matters
+- Limits of unconstrained Black-Box models.
+- Accurate but Physical models.
+- PINNs need less data.
 
-### Part 2: Explainable ML
-- The Black Box problem.
-- Sensitivity analysis (Perturbation theory).
-- Feature importance (SHAP/LIME).
+### Part 2: Automatic Differentiation
+- GradientTape mechanics.
+- Derivatives as ML architecture components.
 
-### Part 3: Causality & Semantics
-- Causal graphs (Cause → Mechanism → Effect).
-- Detection vs. Prediction.
-- Materials Ontologies.
+### Part 3: Solving Physics with NNs
+- PINN Architectures: Data Loss + Physics Loss.
+- Enforcing Boundary Conditions.
+- 3D printing heat transfer case study.
 
-### Part 4: Ethics and Limits
-- Representation and "Success" bias.
-- Dangers of over-extrapolation.
-- Environmental cost vs. PINN efficiency.
+### Part 4: Equation Discovery
+- Sparse Regression and candidate dictionaries.
+- Damped pendulum equation case study.
+- Unit Analysis search pruning.
 
-### Part 5: Final Outlook
-- Self-driving labs.
-- Sustainable AI-driven discovery.
+### Part 5: The Grey-Box Future
+- Hybrid architectures vs. FEA.
+- Building industrial trust.
 
 ## Quarto Website Update (Summary)
-**Summary for ML-PC Week 14:**  
-- Reflects on Explainability and Sensitivity Analysis to open black-box models.
-- Discusses Causality in process chains and Ontologies for semantic reasoning.
-- Critically assesses Data Bias and physical AI hallucinations.
-- Examines the evolving partnership between autonomous algorithms and human experts.
+**Summary for ML-PC Week 13:**  
+- Combines neural networks with physical laws via Physics-Informed ML.
+- Introduces PINNs and automatic differentiation.
+- Details Governing Equation Discovery using sparse regression.
+- Applies physical constraints to build data-efficient Grey-Box models.

week14_summary.md

@@ -1,3 +1,49 @@
-# Week 14 Summary: Buffer, review, and mini-project work
+# Week 14 Summary: Integration, limits, and reflection
 
-No new material. Reserved as a buffer to absorb schedule slippage from the Week 7 / Week 8 public-holiday self-study sessions, for review of difficult topics on request, and for mini-project consultation and presentations.
+## Cross-Book Summary
+
+### 1. Explainability: Opening the Black Box
+- **Beyond Prediction:** "Why" builds industrial trust over just "what".
+- **Sensitivity Analysis:** Perturb inputs to find primary process drivers.
+- **Levels of Explanation:** Tailor explanations for managers, experts, and data scientists.
+
+### 2. Causality and Semantics
+- **Causal Process Chains:** Shift from anomaly detection to early prediction.
+- **Ontologies:** Digitize semantic meaning (e.g., mapping variables to physical concepts) to enable algorithmic reasoning.
+
+### 3. Limits of AI in Materials Science
+- **Data Bias:** Models trained only on "successes" cannot predict failures.
+- **AI Hallucinations:** Large models may generate physically impossible patterns.
+- **Expert's Role:** AI automates analysis, but human experts define questions and interpret final truths.
+
+## 90-Minute Lecture Strategy
+
+### Part 1: Course Synthesis
+- Recap: Signal formation to PINNs.
+- AI-driven materials lifecycle.
+
+### Part 2: Explainable ML
+- The Black Box problem.
+- Sensitivity analysis (Perturbation theory).
+- Feature importance (SHAP/LIME).
+
+### Part 3: Causality & Semantics
+- Causal graphs (Cause → Mechanism → Effect).
+- Detection vs. Prediction.
+- Materials Ontologies.
+
+### Part 4: Ethics and Limits
+- Representation and "Success" bias.
+- Dangers of over-extrapolation.
+- Environmental cost vs. PINN efficiency.
+
+### Part 5: Final Outlook
+- Self-driving labs.
+- Sustainable AI-driven discovery.
+
+## Quarto Website Update (Summary)
+**Summary for ML-PC Week 14:**  
+- Reflects on Explainability and Sensitivity Analysis to open black-box models.
+- Discusses Causality in process chains and Ontologies for semantic reasoning.
+- Critically assesses Data Bias and physical AI hallucinations.
+- Examines the evolving partnership between autonomous algorithms and human experts.