This slide will try to communicate via pictures, instead of going technical mumbo-jumbo. We might go somewhere but slide is full of pictures. If you dont understand any part of it, let me know.
2. Contents
• What is Machine Learning?
• Types of Machine Learning
• Decision Tree and Random Forests
• Neural Network
• Deep Learning
• Forecasting
• Measuring Performance of ML algorithms
• Pitfalls of Machine Learning
3. What is Machine Learning?
• Definition by Arthur Samuel - Machine Learning is a technique which gives
"computers the ability to learn without being explicitly programmed.“
• To mimic human intelligence or human learning process
5. How Humans learn?
• Knowledge Transfer –attending
lecture
• Hit and Trial – learning cycling
during your childhood
What is actually
happening?
• Categorization
• Prediction
7. Supervised Machine learning
• Formal boring definition - Supervised learning task of inferring a function from
labeled training data. The training data consist of a set of training examples. In
supervised learning, each example is a pair consisting of an input object (typically
a vector) and a desired output value (also called the supervisory signal).
• Layman term – Make computers learn from experience
• Task Driven
9. Example of supervised Machine
Learning
Categorization
Categorizing whether tumor is
malignant or benign
Prediction (Regression)
Predicting the house of price in
given area
10. How is Supervised Learning Achieved?
• The existing data is usually divided into
training or testing set. For e.g.,
Training is usually 70% of data where as
Testing is 30% of data.
• Algorithm develops it model based on
training data
• Features important for model is usually
selected by humans
• Algorithm predicts the results for
testing data and later the predicted
value is compared with real value to
give us accuracy.
• Several algorithms are tried until
required accuracy is achieved
11. UnSupervised Learning
Unsupervised learning is a type of machine learning algorithm used to draw
inferences from datasets consisting of input data without labeled responses.
It is data driven.
Clustering a tumor of same kind
but doesn’t know it’s nature.
13. Reinforcement Learning
• Reinforcement learning is a setting where we have a sequential decision
problem. Making a decision now influences what decisions we can make in the
future. A reward function is provided that tells us how “good” certain states are.
• For e.g., Making robot learn against worthy opponent to play table tennis
14. Reinforcement Learning
Characteristics
• No direct training examples – (delayed) rewards later
• Need for exploration of environment or exploitation of environment
• The environment might be stochastic and/or unknown
• The actions of the learner affects future rewards
16. Decision Tree
Decision is a simple representation for
Classifying examples.
Decision tree learning is one of the
most successful techniques for
supervised classification learning.
For e.g., Surviving Titanic is famous first
Machine Learning explanation for
Decision Tree
17. Random Forest
• Random Forest Tree is a Supervised Machine Learning Algorithm Based on Decision
Trees.
• It is Collective Decisions of Different Decision Trees.
• In random forest, there is never a decision tree which have all features of all other
decision trees.
18. Boosting
• Form a large set of simple features
• Initialize weights for training images
• For T rounds
• Normalize the weights
• For available features from the set, train a classifier using a single feature and evaluate
the training error
• Choose the classifier with the lowest error
• Update the weights of the training images: increase if classified wrongly by this
classifier, decrease if correctly
• Form the final strong classifier as the linear combination of the T classifiers
(coefficient larger if training error is small)
19. Neural Network
• a computer system modelled on the human brain and nervous system
31. Forecasting
• It is usually done on time-series data
to predict the future trend
• For e.g., forecasting Stock value
based on historical data
• Usually achieved by ARIMA (Auto-
regressive integrated Moving
Average) model
Amazon stock forecast
Amazon actual stock performance
33. Pitfalls of Machine Learning
• Over fitting
• Trying to make algorithm to work only for small set of data
• Ignoring Human intuition
• Forecasting usually fails because the algorithm is not able to gauge what humans consider
valuable.
• Machine Bias
• For example, a 2015 study found that women were less likely to be shown high-income job ads by
Google's AdSense and another study found that Amazon’s same-day delivery service was
systematically not available in black neighborhoods, both for reasons that the companies could
not explain, but were just the result of the black box methods they used
• https://youtu.be/tleeC-KlsKA