Machine Learning Algorithms
A structured program for practitioners who want to understand how algorithms actually work - not just which library to call.
What this program covers
The curriculum moves from foundational probability and linear algebra through supervised, unsupervised, and reinforcement learning - each module grounded in real datasets.
Nova-agent was built on the premise that an AI for Mac should handle real tasks - research, monitoring, browser work, and automation - so you focus on what matters. This program applies the same logic to learning: structured, purposeful, and directly applicable.
Self-study - or guided instruction?
Both paths lead to the same material. The difference is in how much structure and feedback you need to stay on track.
Self-paced access
Work through recorded sessions and labs on your own schedule. No deadlines, no cohort.
- Full video library with transcripts
- Downloadable lab notebooks
- Community forum access
- Certificate on completion
Cohort program
Eight weeks of live sessions, peer review, and direct feedback from instructors on your submitted work.
- Weekly live sessions with Q&A
- Graded assignments with written feedback
- Peer study groups organized by timezone
- One-on-one office hours each module
- Project review before certification
Curriculum at a glance
Eight modules arranged so each one builds on the last. Participants who skip foundations consistently struggle at the midpoint - the sequence is intentional.
Covers matrix operations, eigendecomposition, Bayes' theorem, and gradient descent from scratch. Participants work through proofs by hand before implementing in NumPy - the goal is understanding, not memorization of API calls.
Decision trees, k-nearest neighbors, logistic regression, and support vector machines. Each algorithm is implemented from scratch, then compared against scikit-learn equivalents to verify correctness and benchmark performance.
K-means, DBSCAN, PCA, and t-SNE applied to real tabular and image datasets. The module spends considerable time on the question of evaluation - how do you measure quality when there are no ground-truth labels?
Builds a small neural network in plain Python before introducing PyTorch. Covers vanishing gradients, batch normalization, dropout, and learning rate schedules - the practical details that make training stable rather than just theoretically sound.