Classification & Regression
Learn mappings from labeled examples and evaluate how well the model generalizes.
JHU / CS475
FALL ARCHIVE
Machine
Learning
ŷ = argmaxy P(y | x, θ)
From probability and linear algebra to classifiers, latent variables and neural models — a course about turning data into decisions.
SUPERVISED
UNSUPERVISED
OPTIMIZATION
MODELS
CODE
475
REASONING
REASONING
01
COURSE OVERVIEW
Learn the model.
Learn the model.
Then learn why it works.
CS475 introduces the foundations of modern machine learning: methods that learn patterns from data and improve predictive performance through statistical structure rather than hard-coded rules.
The course combines mathematical reasoning with substantial programming work, requiring both implementation and analysis.
02 / CORE IDEAS
The vocabulary of learning.
Support Vector Machines
Separate classes with geometry, optimization and maximum-margin reasoning.
Logistic Models
Connect linear predictors with probabilistic outputs and likelihood-based training.
Boosting
Combine weak learners into stronger predictive systems through iterative weighting.
Expectation Maximization
Reason about hidden variables through alternating expectation and optimization.
Neural Learning
Build layered representations and understand how modern models optimize them.
03 / RESEARCH PRACTICE
Models become real when the code runs.
Assignments move ideas off the whiteboard and into a common learning framework where implementation choices, numerical behavior and evaluation all matter.
04
Module map
A conceptual route from mathematical foundations to increasingly expressive models.
W01–02
Foundations
Probability, linear algebra, multivariable calculus and optimization refresh.
W03–04
Linear Models
Loss functions, logistic regression, decision boundaries and generalization.
W05–06
Margins & Kernels
Support vectors, constrained optimization and richer feature spaces.
W07–08
Ensembles
Boosting, combinations of predictors and error-focused iteration.
W09–10
Latent Variables
Expectation Maximization, hidden structure and probabilistic reasoning.
W11–13
Graphical & Deep Models
Structured dependencies, representation learning and modern neural methods.
05 / COURSE WORK
Reason on paper.
Reason on paper.
Implement in code.
Analytical Problems
Derive objectives, inspect assumptions and explain model behavior using mathematical reasoning.
PROBABILITY / LINEAR ALGEBRA / OPTIMIZATIONProgramming
Implement learning algorithms inside a common framework and validate them on data.
CODE / DEBUG / EVALUATELearning System
Bring models and engineering together into an end-to-end final learning project.
DESIGN / TRAIN / TEST
Z
θ*
06 / OPTIMIZATION NOTE
Zeus as the strike that finds the minimum.
Zeus appears here as an editorial metaphor for optimization: energy moving through a parameter landscape until the learning process settles on a solution that explains the data better.
07 / MODEL PIPELINE
Data in.
Data in.
Evidence out.
DATA
Features, labels and observations.
→
MODEL
A structured hypothesis.
→
LOSS
A measure of being wrong.
→
OPTIMIZE
Update parameters.
→
EVALUATE
Test generalization.
08 / RESEARCH CULTURE
A model is an argument made with data.
Machine learning is more than achieving a high score. Good work asks whether the data is appropriate, whether the metric means what we think it means and whether the conclusions hold outside the training set.
09 / LANGUAGE DATASET
DIGITAL CULTURE NOTE
Slot Gacor is a useful example of context changing meaning.
Natural language models learn that words depend on communities and context. Slot Gacor appears here only as an editorial example of a phrase whose meaning comes from a particular online vocabulary rather than from its component words alone.
10 / ARCHIVE
601.475
RETURN TO MODEL ↑
601.475
Machine Learning
INSTITUTIONJOHNS HOPKINS UNIVERSITY
DISCIPLINECOMPUTER SCIENCE / AI
FORMATLECTURE / CODE / ANALYSIS