# ILLINOIS CS 446 - 082917.2 (7 pages)

Previewing pages*1, 2*of 7 page document

**View the full content.**## 082917.2

Previewing pages
*1, 2*
of
actual document.

**View the full content.**View Full Document

## 082917.2

0 0 40 views

- Pages:
- 7
- School:
- University of Illinois - urbana
- Course:
- Cs 446 - Machine Learning

**Unformatted text preview: **

CS446 Machine Learning Fall 2017 Lecture 1 Intro to ML Nearest Neighbor Classifiers Lecturer Sanmi Koyejo Scribe Davis Freimanis Aug 29th 2017 Machine Learning Introduction Basics Key Ideas Machine learning is everywhere And in the beginning of the lecture a few machine learning algorithms were written down from the audience1 presented in Figure 1 Figure 1 Algorithms The way Sanmi organized the given algorithms was by putting unsupervised generative algorithms on the bottom supervised on the top probabilistic on the left and nonprobabilistic on the right The given algorithms doesn t cover all of machine learning but it covers quite 1 Gradient decent was also mentioned but is not an algorithm by itself It is a way to optimize algorithms This method will be covered later in this class 1 2 1 Intro to ML Nearest Neighbor Classifiers a bit This is only a broad categorization of algorithms and there are algorithms in between of these categories Probabilistic algorithms are built around learning some kind of probability distribution as a way to do machine learning while nonprobabilsitic algorithms don t involve learning a distribution directly We will cover probabilistic and nonprobabilistic algorithms in this class quite a bit In supervised algorithms data is presented as D xi yi where xi is the input and yi is its label In unsupervised algorithms data doesn t include labels and is presented as D xi In supervised algorithms the goal is to learn a mapping from inputs x to y given the data set and the goal in unsupervised is to find some kind of structure in the x s There are also semi supervised algorithms where there is a mix of labeled and unlabeled data Reinforcement learning was also mentioned but will not be covered in this class unless we have time in the end There are other classes covering that e g the AI course CS440 Learning In this section we will cover the following learning functions constant memorization and k nearest neighbors We want to build a learner for supervised learning and are given the data D x1 y1 xn yn where each of the items are input output pairs The goal is to learn some function h such that h xn 1 gives an accurate prediction y n 1 Constant function One of the simplest learners to this problem would be a constant function h x c where c is a constant and yi is binary and can hold the value 1 The constant could be chosen as the average or the most popular value in the data set One good property of this function is that it has low variance which means that if it is trained on new data sets then it will give a fairly constant answer but it does not adapt to whatever we try to solve for that reason Memorization A smarter function is memorization and can be represented as follows h x y x if you have seen x before if x is

View Full Document