For practitioners who can code but feel lost when a paper or framework doc turns to vectors, gradients and probability, and want the specific math that modern ML and LLMs rely on.
The curriculum is being prepared and reviewed. Enrollment is not open yet.
About this course
You do not need a math degree to work in machine learning, but you do need a specific toolkit: vectors and matrices for representing data and model weights, derivatives and the chain rule for gradients, probability for model outputs and losses, and optimization for training. This course teaches exactly that toolkit, each idea paired with a short NumPy or PyTorch example so you can check your understanding by running code.
You will work through matrix multiplication and its shapes, dot products and cosine similarity for embeddings, eigenvectors and SVD for dimensionality reduction and low-rank adaptation, derivatives, gradients and backpropagation with PyTorch autograd, probability distributions, softmax, cross-entropy and maximum likelihood, and gradient descent with momentum and Adam. A final module combines these into scaled dot-product attention and a complete training step, and reads the numerical-precision concerns (floating-point formats, overflow, log-sum-exp) that matter on real hardware. The course ends with an automated exam and a Wirewalk Institute certificate of completion and assessment.
What you will be able to do
Manipulate vectors and matrices, predict the shape of matrix products, and implement them with NumPy broadcasting
Use dot products, norms and cosine similarity to reason about embeddings and nearest-neighbor search
Explain eigen-decomposition and singular value decomposition and their use in PCA and low-rank methods such as LoRA
Compute derivatives and gradients with the chain rule and verify them with PyTorch autograd
Apply probability distributions, softmax, cross-entropy and maximum likelihood to classification and language-model losses
Explain gradient descent, learning rates, momentum and Adam, and diagnose a diverging or stalled training curve
Derive scaled dot-product attention step by step and identify numerical-precision pitfalls in training
Syllabus
6 modules · 18 lessons · about 14 hours
1
Module 1: Vectors, Matrices and Shapes
The linear algebra that represents data, weights and embeddings.3 lessons
Vectors, Dot Products and Cosine Similarity
Matrix Multiplication, Shapes and NumPy Broadcasting
Tensors in PyTorch: Batches, Dimensions and Reshaping
2
Module 2: Decompositions and Low Rank
Why many large matrices can be approximated by small ones.3 lessons
Eigenvectors and Eigenvalues by Example
Singular Value Decomposition and PCA
Low-Rank Approximation and Why LoRA Works
3
Module 3: Calculus for Learning
Derivatives, gradients and the chain rule that powers backpropagation.3 lessons
Derivatives and Partial Derivatives as Sensitivity
Gradients and the Chain Rule
Backpropagation and PyTorch Autograd
4
Module 4: Probability and Statistics
How models express uncertainty and how losses are derived.3 lessons
Random Variables, Distributions and Expectation
Softmax, Cross-Entropy and Maximum Likelihood
Bayes' Rule, Sampling and Confidence Intervals for Evaluation
5
Module 5: Optimization
How training actually moves weights toward a better model.3 lessons
Gradient Descent, Learning Rates and Loss Landscapes
Stochastic Mini-Batches, Momentum and Adam
Regularization, Weight Decay and Learning-Rate Schedules
6
Module 6: Putting It Together
Assemble the math into the core computations of modern models.3 lessons
Scaled Dot-Product Attention, Step by Step
Anatomy of One Training Step: Forward, Loss, Backward, Update
Floating-Point Formats, Overflow and the Log-Sum-Exp Trick
Exam and certificate
The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 40 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.
Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.
The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by NumPy or PyTorch. Product and company names are trademarks of their respective owners.
A free, plain-English introduction taught through one realistic company getting ready for its busiest season. Sort the kinds of AI, see why a chat assistant can sound right and be wrong, decide which tasks it fits and how to measure them, catch its three predictable failures, ask well, and decide what never goes into an AI tool.
A plain-language grounding in artificial intelligence: what AI, machine learning and generative AI actually are, how chat assistants produce text, why they sometimes get things confidently wrong, and how to use them safely for real work.
A guided tour of the machinery inside a large language model, told in plain-language analogies, tables and small runnable examples rather than equations: tokenizers, embeddings, attention, pretraining, instruction tuning, reasoning models, context windows and sampling. You leave able to explain why models behave the way they do.