ChoralSoft Learn
Learning paths
Pick a path, start it today, and one new lesson opens every week.
Zero to ML math in 12 weeks
One lesson a week, from probability basics to how modern language models are trained and measured.
Week 1: Averages and the central limit theorem
Why sample averages behave so predictably, and why that matters for everything that follows.
Week 2: Updating beliefs with Bayes’ rule
Start with a prior, add evidence, get a posterior. Play with it in the interactive demo.
Week 3: Maximum likelihood and MAP
Two ways to estimate a hidden quantity from data, and how a prior changes the answer.
Week 4: Cross-entropy and KL divergence
How to measure the distance between what a model believes and what is true.
Week 5: Eigenvalues and symmetric matrices
The linear algebra idea behind PCA, covariance and many stability results.
Week 6: Condition numbers
When tiny input errors become huge output errors, and how to spot it.
Week 7: Optimizers: SGD, momentum and Adam
What each optimizer adds to plain gradient descent.
Week 8: Convex optimization
The well-behaved corner of optimization, using the support vector machine as an example.
Week 9: Bias and variance
Diagnosing whether a model is too simple or too flexible.
Week 10: Measuring models honestly
Every benchmark score is an estimate, so learn to put error bars on it.
Week 11: Attention
Why attention replaced recurrence, and why scores are divided by √d.
Week 12: Scaling laws and perplexity
How model quality is measured and how it improves with compute.