Cursarium logoCursarium
intermediateFree

Introduction to Machine Learning

by Leslie Kaelbling & Tomás Lozano-Pérez · MIT OpenCourseWare

4.6
(1,200 reviews)
80K+ enrolled14 weeksUpdated 2024-01

Our Verdict

Worth it — with caveats

MIT 6.036 Introduction to Machine Learning is the actual on-campus undergraduate ML course (taught by Leslie Kaelbling and Tomas Lozano-Perez, with Fall 2020 video lectures by Tamara Broderick) released free through MIT OpenCourseWare and the MIT Open Learning Library. It is a rigorous, math-and-algorithms-first survey that builds from perceptrons and margin maximization through linear/logistic regression, neural networks and backpropagation, CNNs/RNNs, clustering, and reinforcement learning, with lecture notes, exercises, weekly labs, and homework. This is genuinely intermediate-level material: the official syllabus expects roughly 12 hours per week over 13 weeks, and you need Python plus comfort with calculus and linear algebra to keep up. It rewards learners who want to understand the mathematics behind the methods rather than just call library functions, but the OCW package is less polished than a paid product and has no certificate or graded feedback. For learners who want MIT-grade depth at zero cost and are willing to self-direct, it is an excellent free option.

Outstanding free, MIT-rigor curriculum that is worth taking IF you already have Python, calculus, and linear algebra and want the math/algorithms behind ML. It is the wrong starting point for absolute beginners or people who only want applied, project-first ML, and the OCW version gives you no certificate, no autograder, and no instructor support.

Best for: Self-directed learners and CS/engineering students who want undergraduate MIT-level rigor and prefer working through derivations, proofs, and problem sets; people with a programming and math background who want to understand WHY ML algorithms work (perceptron, SVM margins, backprop, EM, MDPs) rather than just use a library; anyone wanting a free, high-quality theoretical foundation before deep learning courses.

Skip if: Complete beginners with no calculus or linear algebra, learners who want a gentle, hand-held or purely applied/project-based path, people who need a shareable certificate or graded feedback, and those looking for a quick survey. The OCW/Open Learning Library version has no certificate, no monitored discussion forum, and the lecture quality is reported as uneven, so it is hard going without self-discipline and prior math.

About This Course

MIT course covering perceptrons, feature representation, margin maximization, regression, and neural networks foundations.

What You'll Learn

Formulate supervised learning problems and reason about representation, over-fitting, regularization, and generalization
Linear classifiers from first principles: the perceptron algorithm, maximum-margin hyperplanes, hinge loss, and stochastic gradient descent
Linear and logistic regression, plus non-linear classification with feature transformations and kernels
Neural networks and deep learning, including backpropagation, convolutional networks (CNNs), and recurrent networks (RNNs)
Unsupervised learning: clustering, generative/mixture models, and the EM algorithm
Probabilistic models including Hidden Markov Models and Bayesian networks
Reinforcement learning and sequential decision-making with Markov decision processes

Curriculum

Linear classifiers & the perceptron

Introduction to ML; linear classifiers, separability, and the perceptron algorithm.

Margins, loss & regularization

Maximum-margin hyperplanes, loss functions, regularization, stochastic gradient descent, over-fitting, and generalization.

Regression

Linear regression and (in later versions) logistic/linear logistic classification.

Kernels & recommender systems

Non-linear classification via kernels; recommender problems and collaborative filtering.

Neural networks & deep learning

Learning features; feed-forward neural networks; deep learning and backpropagation; convolutional neural networks.

Sequence & state models

Recurrent neural networks; state machines and Markov decision processes.

Unsupervised learning

Generalization/VC-dimension; clustering; generative and mixture models; the EM algorithm; decision trees and random forests.

Probabilistic models

Hidden Markov Models, Bayesian networks, and probabilistic inference.

Reinforcement learning

Learning to control: reinforcement learning and sequential decision-making.

Prerequisites

  • Python programming (e.g., MIT 6.0001/6.00.1x level)
  • Single- and multivariable calculus
  • Linear algebra (vectors and matrices)
  • Basic probability is strongly recommended (helps significantly with the later probabilistic-modeling and EM units)

Instructor

Leslie Kaelbling & Tomás Lozano-Pérez

Instructor · MIT OpenCourseWare

Pros & Cons

Pros

  • Genuine MIT undergraduate course materials (lecture notes, slides, full Fall 2020 video lectures, exercises, labs, and homework) released completely free with no paywall
  • Strong mathematical and algorithmic depth taught from first principles, frequently praised for breadth across the ML landscape in a single course
  • Detailed, well-regarded lecture notes and TA/lab materials that many self-learners cite as the most valuable part
  • Builds durable foundations (perceptron, SVM margins, backprop, EM, MDPs) that transfer directly to more advanced deep-learning and research courses

Cons

  • No certificate, no autograder, and no monitored discussion forum in the free OCW/Open Learning Library version; you self-assess and self-support
  • Steep prerequisites: without solid calculus and linear algebra the material is reported as significantly harder to follow
  • Lecture quality is described by learners as uneven, with some lectures lacking critical detail, leaving occasional gaps you must fill yourself
  • Theory- and proof-heavy with relatively little hand-holding or applied, portfolio-style project guidance compared with commercial courses

Alternatives To Consider

Frequently Asked Questions

Is Introduction to Machine Learning free?

Yes — Introduction to Machine Learning is free to access. Free with no certificate via MIT OpenCourseWare and the MIT Open Learning Library (optional free enrollment lets you track progress). The closely related credentialed path is the MITx edX course 6.86x 'Machine Learning with Python: from Linear Models to Deep Learning' (part of the Statistics and Data Science MicroMasters), which can typically be audited free but charges a fee for the verified certificate.

Who is Introduction to Machine Learning for?

Self-directed learners and CS/engineering students who want undergraduate MIT-level rigor and prefer working through derivations, proofs, and problem sets; people with a programming and math background who want to understand WHY ML algorithms work (perceptron, SVM margins, backprop, EM, MDPs) rather than just use a library; anyone wanting a free, high-quality theoretical foundation before deep learning courses.

What will you learn in Introduction to Machine Learning?

Formulate supervised learning problems and reason about representation, over-fitting, regularization, and generalization; Linear classifiers from first principles: the perceptron algorithm, maximum-margin hyperplanes, hinge loss, and stochastic gradient descent; Linear and logistic regression, plus non-linear classification with feature transformations and kernels; Neural networks and deep learning, including backpropagation, convolutional networks (CNNs), and recurrent networks (RNNs).

What are the prerequisites for Introduction to Machine Learning?

Python programming (e.g., MIT 6.0001/6.00.1x level); Single- and multivariable calculus; Linear algebra (vectors and matrices); Basic probability is strongly recommended (helps significantly with the later probabilistic-modeling and EM units).

Is Introduction to Machine Learning worth it?

Outstanding free, MIT-rigor curriculum that is worth taking IF you already have Python, calculus, and linear algebra and want the math/algorithms behind ML. It is the wrong starting point for absolute beginners or people who only want applied, project-first ML, and the OCW version gives you no certificate, no autograder, and no instructor support.