Getting Started With ML

Beginners Guide to Machine Learning

Well it got redundent over time explaning the same thing to juniors, and I kept on missing out on stuff, so here’s a comprehensive guide on how one should approach Machine Learning.

Table of Content

1. What I Usually Suggest
2. Jump over to a structured topicwise guide inspired by IITK
3. Here’s a simple guide from the ‘Introduction to ML’ session from Jan/Feb 2023, and the session notes as well.

What I Usually Suggest

Assuming you are an eager freshman / sophomore, you should start simply by Visualizing common datasets, which is formally referred to as Exploratory Data Analysis.

• If you have heard of The Prog Club, chances are they’ll have a summer course for this as well, which started as a bootcamp in the summers of 2023, aimed at freshmen transitioning into sophomores.

• Read about eigenvalues, eigenvectors (class 12 topics, for intuition check out a visualization by 3Blue1Brown)

• Read about PRP Probability and Random Processes, I took a course by Prof. SNS in my sophomore year which helped me cement ideas of Random Variables and Expectations of RV.

• Revise basic Probability and Statistics concepts as well, to know about the different distributions out there and some universal laws like the Central Limit Theorem.

Keep experimenting with datasets, find outliers, missing data, basic information about the data from visualizing trends across variables.

Here's a guide inspired by a introductory course to ML from IITK

Preliminaries

Multivariate calculus: gradient, Hessian, Jacobian, chain rule
Linear algebra: determinants, eigenvalues/vectors, SVD
Probability theory: conditional probability, marginal probability, Bayes rule

Supervised Learning

nearest-neighbors, decision trees
Learning by function approximation
Linear models: (multiclass) support vector machines, ridge regression
Non-linear models: kernel methods, neural networks (feedforward)
Learning by probabilistic modeling
Discriminative methods: (multiclass) logistic regression, generalized linear models
Generative methods: naive Bayes

Unsupervised Learning

Discriminative Models: k-means (clustering), PCA (dimensionality reduction)
Generative Models Latent variable models: expectation-maximization for learning latent variable models
Applications: Gaussian mixture models, probabilistic PCA
Concepts of over-fitting and generalization, bias-variance tradeoffs
Optimization for machine learning: (stochastic/mini-batch) gradient descent

How to proceed ?

Ensemble methods: boosting, bagging, random forests
Deep learning: CNN, RNN, LSTM, autoencoders