Wikipendium

Share on Twitter Create compendium Add Language
Edit History
Tools
  • Edit
  • History
  • Share on Twitter

  • Add language

  • Create new compendium
Log in
Table of Contents
  1. Introduction
  2. Course content
    1. Main types of learning methods
      1. Classification
      2. Regression
  3. Reinforcement learning
  4. Unsupervised learning
‹

MICRO-570: Advanced Machine Learning

Tags:
  • Machine learning
+

Introduction

This compendium is made for the course CS-450 Advanced Machine Learning at École Polyteqnique Fédérale de Lausanne (EPFL) and is a summary of the lectures and lecture notes. Is is not the complete curriculum, but rather a list of reading material.

Professor: Aude Billard

Webpage: http://lasa.epfl.ch/teaching/lectures/ML_MSc_Advanced/index.php

Course content

Main types of learning methods

  • Supervised learning
    • Classification
    • Regression
  • Reinforcement learning
  • Unsupervised learning

Classification

Typically binary classification, which can be combine into classifying many classes. Simple example is image classification.

Learning facital expressions can be difficult if we don't preprocess the data to standarize all inputs.

Always split the data set and use one part for training and one part for testing.

Classification is also called supervised clustering process.

One problem is unbalanced classes

  • Support vector machine
  • Relevance vector machine
  • Boosting - random projections
  • Boosting - random gaussians
  • Random forest
  • Gaussian process

Regression

Map N-dim. input $x \in R^N$ to a continuous output $y \in R$. Learn a function og the type; $f:R^N \rightarrow R$ and $y = f(x)$.

Issues: - Influenced on choice of training points

  • Support vector regression
  • Relevance vector regression
  • Gausiiean process regression
  • Gradient boosting
  • Locally weighted projected regression

Reinforcement learning

Unsupervised learning

Observe a series of input: $D = \{x_1, x_2, ... x_M\}, x\in R^N$

Find regularities in the inpute.g.g correlations, patterns.

Use these to reduve the dimentioality of the data.

Pattern Recognition: Determining Trends: Moving time window

Pattern Recognition: Clustering I.e. medical images

Clustering assume groups of points are similar. Can discover when the metric changes according to the same metric of similarity

  • PCA
  • Kernel PCA
  • Genetic algoritms
  • Kernel K-means
  • Isomap
  • Laplacian map
  • Projections on Lapacian eigenvector

Written by

Martin Hallén
Last updated: Thu, 25 Feb 2016 10:59:21 +0100 .
  • Contact
  • Twitter
  • Statistics
  • Report a bug
  • Wikipendium cc-by-sa
Wikipendium is ad-free and costs nothing to use. Please help keep Wikipendium alive by donating today!