Interactive Notebooks - All Tutorials

Interactive Jupyter Notebooks

This page provides a comprehensive overview of all Jupyter notebooks available across the tutorial series. Each notebook opens directly in Google Colab for immediate interactive exploration.

How to use these notebooks:

• 📓 Click “Open in Colab” to launch the notebook in your browser
• ✏️ Run cells, modify code, and experiment with parameters
• 💾 Save a copy to your Google Drive to keep your changes
• 📚 Return to the linked tutorial chapters for detailed explanations

Tutorial 1: Discrete Probability

First GenJAX Model

Notebook: 📓 Open in Colab: first_model.ipynb

What it covers:

• Your first probabilistic model in GenJAX
• Simulating Chibany’s lunch choices (hamburger vs tonkatsu)
• Basic probability calculations with discrete outcomes
• Understanding random sampling and probability distributions

Related Tutorial Chapters:

• Tutorial 1, Chapter 3: Probability from Counting

Topics:

• Discrete probability distributions
• Random sampling
• GenJAX basics
• Probability visualization

Conditioning and Bayes’ Rule

Notebook: 📓 Open in Colab: conditioning.ipynb

What it covers:

• Conditional probability in practice
• Implementing Bayes’ rule in GenJAX
• The taxicab problem with interactive examples
• Sequential belief updating

Related Tutorial Chapters:

• Tutorial 1, Chapter 4: Conditional Probability

Topics:

• Conditional probability
• Bayes’ theorem
• Posterior belief updating
• Prior and likelihood

Bayesian Learning

Notebook: 📓 Open in Colab: bayesian_learning.ipynb

What it covers:

• Complete taxicab problem with visualizations
• Sequential Bayesian updating with multiple observations
• Interactive sliders to explore different base rates and accuracies
• How prior beliefs affect posterior conclusions

Related Tutorial Chapters:

• Tutorial 1, Chapter 5: Bayes’ Theorem
• Tutorial 2 (GenJAX), Chapter 4: Conditioning

Topics:

• Bayesian inference
• Sequential updating
• Base rate effects
• Prior-posterior relationships

Tutorial 2: GenJAX Programming

Your First GenJAX Model (Tutorial 2)

Notebook: 📓 Open in Colab: 02_first_genjax_model.ipynb

What it covers:

• Building generative models in GenJAX
• Interactive widgets to adjust parameters in real-time
• Visualizing probability distributions
• Understanding how parameter changes affect outcomes

Related Tutorial Chapters:

• Tutorial 2 (GenJAX), Chapter 0: Getting Started
• Tutorial 2 (GenJAX), Chapter 2: First Model

Topics:

• GenJAX generative functions
• Parameter exploration
• Interactive visualization
• Model simulation

Tutorial 3: Continuous Probability & Bayesian Learning

Gaussian Bayesian Interactive Exploration

Notebook: 📓 Open in Colab: gaussian_bayesian_interactive_exploration.ipynb

What it covers:

• Part 1: Gaussian-Gaussian Bayesian Updates

  • Interactive sliders to adjust likelihood variance
  • Sequential observation addition with real-time posterior updates
  • Comparison of posterior vs predictive distributions
  • Effect of measurement noise on learning

• Part 2: Gaussian Mixture Categorization

  • How priors affect decision boundaries
  • Effect of variance ratios on categorization
  • Marginal (mixture) distribution visualization
  • Understanding bimodal vs unimodal distributions

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Main Page
• Tutorial 3 (Intro2), Chapter 4: Bayesian Learning
• Tutorial 3 (Intro2), Chapter 5: Mixture Models

Topics:

• Gaussian distributions
• Bayesian parameter learning
• Conjugate priors
• Posterior inference
• Mixture models
• Decision boundaries

Assignment 1: Gaussian Bayesian Update

Notebook: 📓 Open in Colab: solution_1_gaussian_bayesian_update.ipynb

What it covers:

• Part (a): Visualizing the prior distribution
• Part (b): Effect of likelihood variance (σ²_x = 0.25 vs. 4)
• Part (c): Effect of number of observations (N=1 vs. N=5)
• Part (d): Precision-weighted averaging in action
• Part (e): Posterior vs. predictive distributions
• Part (f): GenJAX verification of analytical formulas

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 4: Bayesian Learning

Topics:

• Gaussian conjugate priors
• Likelihood variance effects
• Posterior concentration
• Predictive distributions
• Precision weighting

Assignment 2: Gaussian Clusters

Notebook: 📓 Open in Colab: solution_2_gaussian_clusters.ipynb

What it covers:

• Part (a): Deriving P(category|observation) using Bayes’ rule
• Part (b): Effect of priors on decision boundaries
• Part (c): Effect of variance ratios on categorization
• Part (d): Computing marginal distributions
• Part (e): Understanding bimodal vs. unimodal mixtures
• Part (f): GenJAX simulation of mixture models

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 4: Bayesian Learning - Problem 2
• Tutorial 3 (Intro2), Chapter 5: Mixture Models

Topics:

• Mixture model categorization
• Bayes’ rule with continuous distributions
• Decision boundaries
• Marginal probability
• Mixture distributions

Dirichlet Process Mixture Models (DPMM)

Notebook: 📓 Open in Colab: dpmm_interactive.ipynb

What it covers:

• Interactive DPMM exploration
• Automatic cluster discovery
• Chinese Restaurant Process visualization
• Infinite mixture models
• Bayesian nonparametrics

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 6: DPMM

Topics:

• Dirichlet Process
• Infinite mixture models
• Chinese Restaurant Process
• Bayesian nonparametrics
• Automatic model selection

Bayesian Generalization

Notebook: 📓 Open in Colab: 07_generalization.ipynb

What it covers:

• The sticker warm-up: a concept as a set of hypotheses
• The number game over 1–30 with seven candidate rules
• The size principle under weak vs. strong sampling
• The generalization gradient — predicting which new numbers fit

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 7: Bayesian Generalization

Topics:

• Hypotheses as sets
• The size principle
• Weak vs. strong sampling
• Generalization gradients

Bayesian Networks

Notebook: 📓 Open in Colab: 08_bayes_nets.ipynb

What it covers:

• The Chapter 5 mixture model, re-built explicitly as a Bayes net
• A hierarchical version (a mixing-weight prior stacked on top)
• Chibany’s multi-parent bento network with a conditional probability table
• Inference from an observed weight back to the hidden cluster

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 8: Bayesian Networks

Topics:

• Directed acyclic graphs (DAGs)
• The Markov factorization
• Conditional probability tables
• Ancestral sampling and inference

Conditional Independence & d-Separation

Notebook: 📓 Open in Colab: 09_conditional_independence.ipynb

What it covers:

• The rain / sprinkler / wet-floor collider as a runnable model
• Conditioning on evidence and recovering posteriors by importance sampling
• Watching explaining away happen numerically (0.30 → 0.59 → 0.30)

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 9: Conditional Independence and d-Separation

Topics:

• Chain, fork, and collider patterns
• d-separation
• The Markov blanket
• Explaining away

Causal Bayes Nets & the Do-Operator

Notebook: 📓 Open in Colab: 10_causal_bayes_nets.ipynb

What it covers:

• The smoking / teeth / cancer network as observational vs. interventional models
• Computing P(cancer | teeth) vs. P(cancer | do(teeth)) by Monte Carlo
• Seeing the see/do gap (≈0.098 vs. 0.052) emerge from graph surgery

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 10: Causal Bayes Nets and the Do-Operator

Topics:

• The do-operator and graph surgery
• Confounders
• Observational vs. interventional distributions
• Pearl’s ladder of causation

Information Theory

Notebook: 📓 Open in Colab: 11_information_theory.ipynb

What it covers:

• Estimating entropy and mutual information by Monte Carlo
• The collider creating mutual information from nothing
• I(rain; tea) = 0 jumping to I(rain; tea | sign) ≈ 0.46 bits — explaining away, in bits

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 11: Information Theory

Topics:

• Surprise and entropy
• Mutual information
• Independence in information units
• The collider, in bits

Markov Chains

Notebook: 📓 Open in Colab: 13_markov_chains.ipynb

What it covers:

• Chibany’s bento chain as a transition matrix; sampling sequences from it
• Power iteration converging to the 70/30 stationary distribution from any start
• The stationary distribution as the eigenvalue-1 eigenvector
• A three-state worked example

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 13: Markov Chains

Topics:

• The Markov property and transition matrices
• Stationary distributions and power iteration
• Ergodicity
• GenJAX sequence sampling + jax.lax.scan

Random Walks on Networks

Notebook: 📓 Open in Colab: 14_random_walks_networks.ipynb

What it covers:

• Chibany’s animal network as an adjacency matrix, row-normalized to a transition matrix
• The stationary distribution of a random walk: π ∝ degree (Cat the bridge wins)
• Visit frequency by simulation, matching the degree law
• Hand-rolled PageRank with the ε-teleport on a tiny directed web

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 14: Random Walks on Networks

Topics:

• Graphs, adjacency matrices, degree
• Random walk as a Markov chain on nodes
• π ∝ degree and where it breaks (directed graphs)
• PageRank

Memory Search

Notebook: 📓 Open in Colab: 15_memory_search.ipynb

What it covers:

• A censored random walk on a small semantic network
• The censoring function (report each animal on first visit) and inter-item response times
• The position-1-slowest “switch cost” signature, with no switch rule
• A simulation-based (ABC) sketch that recovers block structure from fluency lists

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 15: Memory Search as a Random Walk

Topics:

• Semantic fluency and clustering/switching
• Censoring; first-hitting times; IRT
• Recovering the human optimal-foraging curve from one process
• Inverting the walk (U-INVITE, SNAFU) and simulation-based inference

Monte Carlo

Notebook: 📓 Open in Colab: 16_monte_carlo.ipynb

What it covers:

• The Monte Carlo estimator: averaging a die roll to 3.5, estimating π by throwing darts
• Rejection sampling and the indicator function; importance sampling with weights $w = p/q$
• Self-normalized importance sampling (unnormalized posteriors) and the effective sample size
• GenJAX importance sampling via model.importance

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 16: Monte Carlo

Topics:

• Estimating expectations and probabilities by sampling; the $1/\sqrt{n}$ rate
• Rejection sampling, inverse-CDF, importance sampling
• Effective sample size as a proposal-quality diagnostic

Particle Filtering

Notebook: 📓 Open in Colab: 17_particle_filtering.ipynb

What it covers:

• Tracking Chibany down a corridor from noisy sensor pings
• The particle filter loop — weight, resample, propagate — and what each step does
• Weight degeneracy and why resampling is the cure (ESS collapsing without it)
• The GenJAX motion model as the propagate step

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 17: Particle Filtering

Topics:

• State-space models; sequential importance sampling
• Weight → resample → propagate; degeneracy
• Particle filters as a process model of human inference

Markov Chain Monte Carlo

Notebook: 📓 Open in Colab: 18_markov_chain_monte_carlo.ipynb

What it covers:

• Metropolis–Hastings on a bimodal target; why the normalizer cancels
• Gibbs sampling on a correlated 2-D Gaussian (always accepts)
• Mixing, burn-in, and the multimodal trap (a trapped chain from two starts)
• Assembling an MH step in GenJAX from the assess scoring primitive

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 18: Markov Chain Monte Carlo

Topics:

• Designing a chain to hit a target; detailed balance
• Metropolis–Hastings and Gibbs sampling
• Burn-in, mixing, and multimodal targets

Sampling the Mind

Notebook: 📓 Open in Colab: 19_sampling_the_mind.ipynb

What it covers:

• MCMC with People: a person’s choices as the Metropolis accept step
• A hybrid Gibbs–Metropolis sampler for the hierarchical Beta-Binomial (bento shops)
• The Beta-Binomial marginal (collapsing θ out) and the mean/concentration reparametrization
• Reading off a predictive for a brand-new, unseen unit

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 19: Sampling the Mind

Topics:

• MCMC with People; recovering a prior from choices
• Hybrid Gibbs + Metropolis on a hierarchical model
• Beta-Binomial conjugacy and the collapsed sampler

Statistical Decision Theory

Notebook: 📓 Open in Colab: 20_statistical_decision_theory.ipynb

What it covers:

• From beliefs to actions: the decision problem, loss, and risk
• Bayes vs minimax criteria, disagreeing on the same numbers
• The loss → estimator rule: 0–1 → mode (MAP), squared → mean, absolute → median
• Expected loss by sampling the posterior in GenJAX
• One and Done: why optimal decisions can come from a single sample (Vul et al.)

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 20: Statistical Decision Theory

Topics:

• Decision theory, loss functions, risk, Bayes vs minimax
• MAP / mean / median as Bayes estimators
• Monte-Carlo decisions and probability matching

Markov Decision Processes

Notebook: 📓 Open in Colab: 21_markov_decision_processes.ipynb

What it covers:

• Building an MDP: Markov chain + reward + a choice of transition matrix
• The transition as a GenJAX generative model
• Value, the Bellman equation, and value iteration on the Chibany MDP
• The discount factor γ and the policy-flip threshold
• Estimating a value by simulating rollouts (the Monte-Carlo bridge)

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 21: Markov Decision Processes

Topics:

• MDPs, policies, the Bellman equation, value iteration
• Discounting and the γ-flip
• Planning by simulation

Q-Learning

Notebook: 📓 Open in Colab: 22_q_learning.ipynb

What it covers:

• The model-free TD update: learning rate α and the TD error
• Q-learning on the GardenPath gridworld
• Reward shaping and positive cycles; potential-based shaping as the fix
• Dyna and MCTS: planning with a learned model (simulation-based RL)
• Reward hacking → RLHF; the TD error → dopamine

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 22: Q-Learning

Inverse RL: Reading Goals from Behavior

Notebook: 📓 Open in Colab: 23_inverse_rl_goal_inference.ipynb

What it covers:

• Goal inference as Bayes’ rule with a softmax policy as the likelihood
• The GenJAX observer: posterior over goals, and the freeze-frame curve
• Ill-posedness, and how the prior + rationality β resolve it
• Theory of Mind = inverse RL (Baker & Tenenbaum)
• IRL at scale: MaxEnt, GAIL, AIRL

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 23: Inverse RL

POMDPs and Belief: Inferring a Hidden World

Notebook: 📓 Open in Colab: 24_pomdps_belief_inference.ipynb

What it covers:

• Belief states b(s) and the Tiger belief update in GenJAX
• α-vectors and the decision threshold (open vs. listen)
• A POMDP is an MDP over beliefs
• Bayesian Theory of Mind (belief + desire)
• Teaching, legibility, and Cooperative IRL

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 24: POMDPs

Modern RL: Preferences, World Models, and Machine Minds

Notebook: 📓 Open in Colab: 25_modern_rl_world_models.ipynb

What it covers:

• RLHF/DPO as preference-based inverse RL (Bradley–Terry reward model in GenJAX)
• Recovering a reward from pairwise preferences; the additive-constant ambiguity
• Amortized vs. Bayesian Theory of Mind (ToMnet)
• World models (MuZero, Dreamer)
• The skeptical LLM Theory-of-Mind debate

Related Tutorial Chapters:

• Tutorial 3 (Intro2), Chapter 25: Modern RL

From Bentos to Vectors: the Linear Algebra You Need

Notebook: 📓 Open in Colab: vectors_and_spaces.ipynb

What it covers:

• Data → vectors: measurements as components, dimension, points in a plane
• Dot product and cosine similarity, fully worked (kind vs. size)
• 4-D vectors and embedding arithmetic (king − man + woman ≈ queen)
• Matrix–vector multiplication as “a dot product per row”; columns = where the basis arrows land
• Composition (two machines collapse into one) and a first ReLU layer

Related Tutorial Chapters:

• Part VIII, From Bentos to Vectors

Capstone — From Rules to Weights

Notebook: 📓 Open in Colab: capstone_from_rules_to_weights.ipynb

What it covers:

• A three-way race on the bento data as data grows: ridge on hand features, a 2-4-1 learned-feature net, and RBF kernel ridge
• Fixed features win when data is scarce; learned/kernel methods overtake as it grows
• Optional stretch: the Belkin et al. (2019) double-descent curve

Related Tutorial Chapters:

• Part VIII, From Rules to Weights

Capstone — Neural Network Fundamentals

Notebook: 📓 Open in Colab: capstone_neural_net_fundamentals.ipynb

What it covers:

• Part A: replicate Minsky & Papert — a linear unit fails XOR, a hidden layer solves it
• Part B: the same 2-layer net on a real task (sklearn digits, 0 vs. 1)
• Anchor: Minsky & Papert (1969) + Rumelhart, Hinton & Williams (1986)

Related Tutorial Chapters:

• Part VIII, Neural Network Fundamentals

Capstone — Transformers & Attention

Notebook: 📓 Open in Colab: capstone_transformers_attention.ipynb

What it covers:

• Build single-head attention from scratch; hand-set keys/queries for a “look at the previous token” head
• Then a mini induction pattern; visualize the weight matrices as heatmaps
• Anchor: Vaswani et al. (2017) + Olsson et al. (2022) induction heads

Related Tutorial Chapters:

• Part VIII, Transformers & Attention

Capstone — LLMs & In-Context Learning (the flagship)

Notebook: 📓 Open in Colab: capstone_llms_in_context_learning.ipynb

What it covers:

• Part A: the exact Bayesian posterior-predictive of a latent-mixture model vs. a genuinely-trained predictor — watch the posterior sharpen with examples (Xie-style)
• Part B: implement the Falck-style martingale test and run it on both — the exact Bayesian passes; is your trained model’s gap noise or systematic drift?
• Part C (stretch): sketch the same test on a real LLM
• Runs both sides of the ICL-is-Bayesian debate

Related Tutorial Chapters:

• Part VIII, LLMs & In-Context Learning

Capstone — World Models & Imagination

Notebook: 📓 Open in Colab: capstone_world_models_imagination.ipynb

What it covers:

• Part A: fit a dynamics model on a gridworld from random-policy data
• Part B: plan by imagination (random-shooting rollouts) and race it against model-free Q-learning on sample efficiency
• Part C (stretch): inject model error and watch planning degrade
• Anchor: Dyna (Sutton 1991) + MuZero (Schrittwieser 2020)

Related Tutorial Chapters:

• Part VIII, World Models & Imagination

Capstone — Adversarial Examples

Notebook: 📓 Open in Colab: capstone_adversarial_examples.ipynb

What it covers:

• Part A: implement the fast gradient sign method (FGSM) and flip a confident classifier
• Part B: sweep the perturbation size ε and plot the robustness curve
• Part C (stretch): a pass of adversarial training, then re-measure
• Anchor: Goodfellow, Shlens & Szegedy (2015) + Szegedy et al. (2014)

Related Tutorial Chapters:

• Part IX, Adversarial Examples

Capstone — Fairness, Formally

Notebook: 📓 Open in Colab: capstone_fairness_formalisms.ipynb

What it covers:

• Part A: implement demographic parity, equalized odds, and calibration on a confusion matrix
• Part B: reproduce the impossibility result on a COMPAS-like dataset with unequal base rates
• Part C (stretch): the fairness frontier (the tradeoff curve)
• Anchor: Kleinberg et al. (2016) + Chouldechova (2017) + the ProPublica/COMPAS case

Related Tutorial Chapters:

• Part IX, Fairness, Formally

Capstone — Bias in Data

Notebook: 📓 Open in Colab: capstone_bias_in_data.ipynb

What it covers:

• Part A: implement WEAT on a toy embedding and measure the effect size
• Part B: attempt linear debiasing and measure what’s removed vs. what leaks back
• Part C (stretch): a downstream task inheriting the bias
• Anchor: Caliskan et al. (2017) + Bolukbasi et al. (2016) + Buolamwini & Gebru (2018)

Related Tutorial Chapters:

• Part IX, Bias in Data

Capstone — Alignment & Safety (the book’s last)

Notebook: 📓 Open in Colab: capstone_alignment_safety.ipynb

What it covers:

• Part A: fit a Bradley–Terry reward model from pairwise preferences (recover a latent reward)
• Part B: demonstrate reward hacking — optimize against the learned reward and watch it exploit a gap
• Part C (stretch): more diverse preferences shrink the ill-posedness
• Anchor: Christiano et al. (2017) + Hadfield-Menell et al. (2016) + the Goodhart/specification-gaming literature

Related Tutorial Chapters:

• Part IX, Alignment & Safety

Topics:

• Q-learning, TD error, ε-greedy exploration
• Reward shaping, positive cycles, reward hacking
• Dyna, Monte Carlo Tree Search, simulation-based RL

Recommended Learning Paths

Path 1: Complete Beginner

1. first_model.ipynb - Start here for GenJAX basics
2. conditioning.ipynb - Learn conditional probability
3. bayesian_learning.ipynb - Master Bayesian updating
4. 02_first_genjax_model.ipynb - Build your first full model
5. gaussian_bayesian_interactive_exploration.ipynb - Explore continuous probability
6. solution_1_gaussian_bayesian_update.ipynb - Practice Gaussian inference
7. solution_2_gaussian_clusters.ipynb - Learn mixture models
8. dpmm_interactive.ipynb - Advanced: infinite mixtures

Path 2: Bayesian Learning Focus

1. bayesian_learning.ipynb - Discrete Bayes’ rule
2. gaussian_bayesian_interactive_exploration.ipynb - Continuous Bayesian inference
3. solution_1_gaussian_bayesian_update.ipynb - Gaussian conjugate priors
4. solution_2_gaussian_clusters.ipynb - Mixture model inference
5. dpmm_interactive.ipynb - Bayesian nonparametrics

Path 3: Quick Interactive Tour

1. 02_first_genjax_model.ipynb - Interactive parameter exploration
2. gaussian_bayesian_interactive_exploration.ipynb - Bayesian learning with sliders
3. dpmm_interactive.ipynb - Automatic clustering

Path 4: Graphical Models & Causality

1. 08_bayes_nets.ipynb - Draw models as graphs
2. 09_conditional_independence.ipynb - d-separation and explaining away
3. 10_causal_bayes_nets.ipynb - Seeing vs. doing (the do-operator)
4. 11_information_theory.ipynb - Measure dependence in bits

Path 5: Markov Chains, Networks & Memory

1. 13_markov_chains.ipynb - Transition matrices and the stationary distribution
2. 14_random_walks_networks.ipynb - Random walks on graphs, π ∝ degree, PageRank
3. 15_memory_search.ipynb - Recall as a censored random walk on a semantic network

Path 6: Sampling & Monte Carlo

1. 16_monte_carlo.ipynb - Estimating by sampling; importance sampling and effective sample size
2. 17_particle_filtering.ipynb - Streaming inference; weight → resample → propagate
3. 18_markov_chain_monte_carlo.ipynb - Designing a chain to hit a target; Metropolis–Hastings and Gibbs
4. 19_sampling_the_mind.ipynb - MCMC with People and a sampler for the hierarchical Beta-Binomial

Path 7: Decisions & Reinforcement Learning

1. 20_statistical_decision_theory.ipynb - From beliefs to actions: loss, risk, Bayes vs minimax, and one-and-done decisions
2. 21_markov_decision_processes.ipynb - MDPs, the Bellman equation, value iteration, and planning by simulation
3. 22_q_learning.ipynb - Model-free Q-learning, reward shaping and positive cycles, Dyna and MCTS
4. 23_inverse_rl_goal_inference.ipynb - Inverse RL: goal inference, softmax likelihood, ill-posedness, ToM = IRL
5. 24_pomdps_belief_inference.ipynb - POMDPs, belief updates, α-vectors, Bayesian ToM, teaching & legibility
6. 25_modern_rl_world_models.ipynb - RLHF/DPO as inverse RL, world models, the LLM theory-of-mind debate
7. vectors_and_spaces.ipynb - Linear algebra from zero: vectors, dot products, matrices as transformations (the Part VIII on-ramp)
8. capstone_from_rules_to_weights.ipynb - Capstone: ridge vs learned features vs kernel ridge as data grows
9. capstone_neural_net_fundamentals.ipynb - Capstone: XOR from Minsky & Papert to a real digit classifier
10. capstone_transformers_attention.ipynb - Capstone: build attention from scratch; previous-token and induction heads
11. capstone_llms_in_context_learning.ipynb - Capstone: run both sides of the ICL-is-Bayesian debate (Xie/Ye vs Falck)
12. capstone_world_models_imagination.ipynb - Capstone: learn a world model, plan by imagination vs model-free Q-learning
13. capstone_adversarial_examples.ipynb - Capstone: implement FGSM, flip a classifier, plot the robustness curve
14. capstone_fairness_formalisms.ipynb - Capstone: the three fairness criteria + the impossibility result as code
15. capstone_bias_in_data.ipynb - Capstone: WEAT on embeddings, linear debiasing and its leaks
16. capstone_alignment_safety.ipynb - Capstone: fit a Bradley-Terry reward, watch it get hacked

Tips for Using Notebooks

Getting Started:

• Click “Open in Colab” to launch any notebook
• Run cells in order (Shift+Enter) to execute code
• Experiment by changing parameter values

Interactive Widgets:

• Many notebooks include sliders and controls
• Adjust parameters and see results update in real-time
• Try extreme values to understand edge cases

Saving Your Work:

• File → Save a copy in Drive (saves to your Google Drive)
• Your experiments and notes will be preserved
• You can share your modified notebooks with others

Troubleshooting:

• If code doesn’t run, try Runtime → Restart runtime
• Make sure to run cells in order from top to bottom
• Check that all required packages are installed (usually automatic in Colab)

All Notebooks at a Glance

Need help? Return to the main tutorial page or consult the glossary for term definitions.

Enjoying the notebooks? This tutorial series is generously funded by the [Japanese Probabilistic Computing Consortium Association (JPCCA)](https://jpcca.org/).