Scribd
Upload
24 views

Lecture 3: Applications of Machine Learning Algorithms Jul. 06 & 09, 2018

This lecture discusses applications of machine learning algorithms, including supervised learning techniques like classification and regression. Classification can be used for tasks like document classification, spam filtering, and image recognition. Regression has applications in stock price prediction, age prediction from video views, and robot arm location prediction. Unsupervised learning is also discussed for applications involving clustering, dimensionality reduction, natural language processing, graph representations, and matrix completion. The key challenges of supervised learning are defining the mapping function between inputs and outputs when the relationship is complex or ambiguous.

Uploaded by

Akash Gupta
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
24 views

Lecture 3: Applications of Machine Learning Algorithms Jul. 06 & 09, 2018

This lecture discusses applications of machine learning algorithms, including supervised learning techniques like classification and regression. Classification can be used for tasks like document classification, spam filtering, and image recognition. Regression has applications in stock price prediction, age prediction from video views, and robot arm location prediction. Unsupervised learning is also discussed for applications involving clustering, dimensionality reduction, natural language processing, graph representations, and matrix completion. The key challenges of supervised learning are defining the mapping function between inputs and outputs when the relationship is complex or ambiguous.

Uploaded by

Akash Gupta
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Lecture 3: Applications of Machine learning algorithms

Jul. 06th & 09th , 2018


Lecturer: Dr Anitha M

1 Supervised Learning
This section will help you to understand what you can expect from this course. It also lists the real worl
applications in order to feel the depth of the subject. As we know that supervised learning can be done based
on classification method or by Regression method. We shall in detail discuss them here:

1.1 Classification
The problem under supervised learning is categorical then it comes under classification problem. Here, the
goal is to map the input variable to its output y, where, y ∈ {1,2,...,C} C is the output class. If C = 2
then it is called as Binary classification and if C > 2 then it is multiclass classification. Labels are used to
determine the classes.

But sometimes the labels may get confused and will not be able to determine the exact class. i.e., there is
an overlap among the classes which results in ambiguity. This type of classification is called as multilabel
classification.
Everything looks simple here. Is it the reality of supervised learning? The answer is a big no.

Question 1:
What is the challenge/problem?
The problem of supervised learning is to formalize or find out the function that makes predictions for the
new data input.
This formalizing is known as function approximation. We know that generally the supervised learning can
be written as y = f (x). But now f is an unknown function. We need to formalize or estimate it in order to
make the predictions on inputs that were not seen before. Hence, we can rewrite the model as fˆ(x).
The diagram below explains the challenge:

There are two sets available yes, no or 0,1. There are two tables presented in the diagram. One table
represents the relation between the classes and features with N × D dimension. Classes and features can be

1
of any number which are critical parameters to be considered for prediction. Another table shows the labels
for the corresponding classes in the previous table. It consists of 1 and 0.
There are some test cases shown in the diagram:
• Blue crescent
• Yellow torus
• Blue arrow
But then these test cases do not fall under the labels shown. Neither it can be sorted based on the shape
nor it can be done using the color. Hence it creates an ambiguity. To remove this ambiguity Probabilistic
Prediction can be used.

1.1.1 Probabilistic Prediction


Linear Algebra and the probability theory are the most important tools of Machine Learning (ML). Prob-
ability theory provides a consistent framework for the quantification and manipulation of uncertainty and
allows us to make optimal predictions given all the information available to us, even though that information
may be incomplete or ambiguous.
There are two forms of probability the ML algorithms use:
• Conditional Probability
• Un-conditional Probability

1.2 Conditional Probability


Conditional probability is a measure of the probability of an event given that another event has occurred.
Conditional Probability is written as P (y|x). i.e., event ’y’ happens if the event ’x’ has happened. Super-
vised learning approach uses this probability since the approach is finding the mapping function or the target
function for mapping the input ’x’ to output class ’y’.
It is represented as P (y|x, D), where, x is the input that needs to be classified and D is the training data
set and y is the output set of class. Conditions are given on the right side of the conditioning bar |. If
we are specific and know the model ’M’ that needs to be used can also be included in the conditions like
P (y|x, D, M ).
Any supervised algorithm the goal is to best estimate the mapping function (f) for the output variable Y
given the input data ’x’. The mapping function is also called as target function. As we need to find the best
C
guess for the input data it is written as: ŷ = fˆ(x) = argmax p(y = c|x, D).
c=1

Question 1:
What is argmax?

2 Real World Applications


1 This section will list you some of the real world applications of ML.
1
Discussed in class and also you can follow the text book by the author Kevin Murphy for images

2
2.1 Supervised Learning - Classification Approach:
• Document classification - which category does the document belong to?

• E-mail spam filtering - Spam or ham?

• Image classification & Hand writing recognition

• Face detection & recognition

2.2 Regression:
• Predicting tomorrow’s stock market

• Predicting the age of a viewer watching a given video on youtube

• Predicting the location of a robot’s arm in 3d space

• predicting the temperature at any location inside a building

2.3 Unsupervised Learning Approach:


• Discovering Clustering - astronomy, e-commerce and biology

• Discovering latent factors - Dimensionality reduction (3d to 2d) and Principle Component Analysis
(PCA).

• Natural Language Processing (NLP)

• Discovering Graph Structure - sparse graph representation is used

• Matrix Completion - Filling the missing data with NaN - ”Not a Number” - eg:- Image inpainting.
Collaborative Filtering and Market Basket Analysis.

You might also like