Lecture on AI and Machine Learning

© Copyright National University of Singapore. All Rights Reserved.
Lecture 11
Machine Learning and AI
CN 5111
Lecturer: Dr Xiaonan Wang
Assistant Professor
Department of Chemical and Biomolecular Engineering
National University of Singapore
NUS Faculty of Engineering, Block E5, Unit #03-04
+65 6601 6221 (Tel) chewxia@nus.edu.sg
1

Outline
•  Recap
•  An Overview of Machine Learning
Applications
Learning system model
Methods and Algorithms
•  Deep Learning
what and how
2

© Copyright National University of Singapore. All Rights Reserved. 3
Stochastic programming is mathematical (i.e. linear, integer, mixed-integer, nonlinear)
programming but with a stochastic element present in the data:
•  in deterministic mathematical programming the data (coefficients) are known
numbers
•  in stochastic programming these numbers are unknown, instead we may have a
probability distribution present.
Key concept:
– Have multiple guesses at the future rather than just one
– Each represented by a scenario s
• Which is a particular “realisation” of the uncertain parameters
• Which has a particular probability Ps
– Divide variables into
• Here and now – what must be decided now and is scenario-independent
• Wait and see – what can be decided when the uncertain parameters become
“revealed”; scenario-dependent
Recap: Stochastic programming
1ss
P =∑

Newsboy Stochastic Programming
•  Purchase x : here and now variable
•  Sales: wait and see
•  Data:
– No of scenarios, s
– Demand (s), Probability (s) follows a distribution (e.g. from past data)
– Buying price $1.0, Selling price is $1.4
Objective: Expected profit
( )s ss
Max z Sales SalesPrice Probability Purchase PurchasePrice= × × − ×∑
• Variables
– Purchase
– Sales (s)
• Constraints
Sales (s) ≤ Purchase , for every s
Sales (s) ≤ Demand (s) , for every s

The search for the global optimum of a multi-dimensional function is
a difficult task!
Derivative optimization methods (Conjugate gradient method or
simplex method) converge to nearest local minimum.
Exact/deterministic methods
–Grid search –Branch and bound –Interval arithmetic –Multi-start
Heuristic/stochastic/random search methods
–Simulated annealing (SA)
–Genetic algorithms (GA)
–Scatter search
–Tabu search
– Particle swarm optimization (PSO)
Recap: Global optimization
Nature-Inspired
Programming Recipes

Recap: Simulated Annealing (SA)
1.  Start at a given state s0 (point in multi-dimensional parameter space).
Compute cost or energy E(s0). Set temperature to T.
2. Consider somewhat different state s1. Corresponding cost or energy is E(s1).
3. If E(s1) < E(s0), switch from state s0 to state s1. Otherwise: compute
transition probability. Compute random number p uniformly distributed in
interval [0,1]. If p < P ( ) switch to new state s1. Otherwise, stay at s0.
4. Go to 2. (Lower temperature T after a number of steps have been made.)
Remarks: If temperature T is not too small, one cannot get trapped in a local
minimum.
The probability is:
– Proportional to the energy difference
– Modulated through a temperature parameter T
• for T→∞ the probability of any move approaches 1
• for T→0 the probability of any move approaches 0
0 1( )/
0 1( ) E E T
P s s e− −
→ =
0 1s s→

Example of SA and the traveling
salesmen problem
Find a tour of a given set of cities
so that – each city is visited only
once – the total distance
travelled is minimized
https://en.wikipedia.org/wiki/Travelling_salesman_problem

Recap: Genetic Algorithm
An individual is characterized by a set of parameters (variables)
known as Genes. Genes are joined into a string to form
a Chromosome (A string of “Genes” that represents a solution).
The fitness function determines
how fit an individual is.
Select the fittest individuals and
let them pass their genes to the
next generation.
Offspring created
by exchanging the
genes of parents
among themselves
until the crossover
point is reached
Some of genes can be
subjected to a mutation with
a low random probability.
The algorithm terminates if the population has converged (does not produce
offspring which are significantly different from the previous generation).
https://towardsdatascience.com/introduction-to-genetic-algorithms-including-example-code-e396e98d8bf3

Recap: Genetic Algorithm
Matlab Global Optimization Toolbox Overview
https://www.mathworks.com/products/global-optimization.html

Machine Learning
10

Machine Learning
- Grew out of work in AI
- New capability for computers
Examples:
- Database mining
Large datasets from growth of automation/web.
E.g., Web click data, medical records, biology, engineering
- Applications can’t program by hand.
E.g., Autonomous helicopter, handwriting recognition, most of
Natural Language Processing (NLP), Computer Vision.
- Self-customizing programs
E.g., Amazon, Netflix product recommendations
- Understanding human learning (brain, real AI).
Machine Learning
The ML MOOC and deeplearning.ai’s Deep Learning Specialization
teach a lot of technical algorithms by Andrew Ng.

•  Arthur Samuel (1959). Machine Learning: Field of study that gives
computers the ability to learn without being explicitly programmed.
•  Tom Mitchell (1998). Well-posed Learning Problem:
A computer program is said to learn from experience E with respect
to some task T and some performance measure P, if its performance
on T, as measured by P, improves with experience E.
https://www.coursera.org/learn/machine-learning
•  A branch of artificial intelligence, concerned with the design and
development of algorithms that allow computers to evolve
behaviors based on empirical data.
“If you invent a breakthrough in
artificial intelligence, so
machines can learn, that is
worth 10 Microsofts”
(Bill Gates, Chairman, Microsoft)
Machine Learning definition

Machine Learning application
http://blogs.teradata.com/data-points/tree-machine-learning-algorithms/

•  AI automates the monotonous tasks.
•  AI recognizes subtle patterns in well data and
helps predict when each well will need servicing
long before a problem occurs.
•  AI oil and gas systems gradually understand
each operational element better, allowing it to
quickly identify patterns
https://jobs.exxonmobil.com/ExxonMobil/job/Annandale-Machine-Learning-NJ-08801/405778200/

https://www.micron.com/about/blogs/2017/october/machine-learning
1. Modularize Access for Effective Data
Pipelines
2. Create Efficient ML Model Delivery
Strategies
3. Deliver Scalable Computer
Infrastructures

Machine Learning Process
There are several factors affecting the performance:
–  Types of training provided
–  The form and extent of any initial background knowledge
–  The type of feedback provided
–  The learning algorithms used
–  The performance criteria
off line or on line measures, descriptive or predictive output, accuracy, efficiency etc.

Machine Learning models
•  A model of learning is fundamental in any
machine learning application:
·  who is learning (a computer program)
·  what is learned (a domain)
·  from what the learner is learning (the information
source)
•  Training is the process of making the system
able to learn.
Input
Samples
Learning
Method
System
Training
Testing

Machine Learning algorithms
•  Supervised learning
§  Prediction
§  Classification (discrete labels),
§  Regression (real values)
•  Unsupervised learning
§  Clustering
§  Probability distribution estimation
§  Finding association (in features)
§  Dimension reduction
•  Semi-supervised learning
•  Reinforcement learning
Decision making (robot, chess machine)

Supervised learning
–  Linear classifier (numerical functions)
–  Linear/ logistic regression
–  Parametric (Probabilistic functions)
Naïve Bayes, Gaussian discriminant analysis (GDA), Hidden Markov
models (HMM), Probabilistic graphical models
–  Non-parametric (Instance-based functions)
K-nearest neighbors, Kernel regression, Kernel density estimation,
Local regression
–  Non-metric (Symbolic functions)
Classification and regression tree (CART), decision tree
–  Aggregation/ Ensemble learning
Bagging (bootstrap + aggregation), Boosting (Adaboost, Gradient tree
boosting), Random forest
–  Support vector machines
–  Neural networks
–  …
Machine Learning methods

Artificial Neural Networks (ANN)
Models of the brain and nervous system
Highly parallel: Process information much more like the brain
than a serial computer learning
Artificial Neural Networks (ANNs) incorporate the two fundamental
components of biological neural nets:
1. Neurones (nodes)
2. Synapses (weights)
Learning = learning by adaptation
At the neural level the learning
happens by changing of the synaptic
strengths, eliminating some synapses,
and building new ones.

activation function
x2
xn
w1j
w2j
wnj
j
n
i
iijj bxws += ∑=0
)( jj sfy =
yj
bj
js
e
−
+1
1
Artificial Neuron (Linear Units)

S = (1 ! 0.25) + (0.5 ! (-1.5)) = 0.25 + (-0.75) = -
0.5
0.3775
1
1
5.0
=
+ e
Squashing: y =
Sigmoid function js
e
−
+1
1

Recurrent Neural Networks
Feedforward networks:
Information only flows one way
One input pattern produces one output
No sense of time (or memory of previous state)
Recurrent Networks:
Nodes connect back to other nodes or themselves
Information flow is multidirectional
Sense of time and memory of previous state(s)
State-of-the-art results in
time series prediction,
adaptive robotics,
handwriting recognition,
image classification,
speech recognition,
stock market prediction,
and other sequence
learning problems.
Everything
can be
processed
sequentially.

Unsupervised learning
–  Clustering
•  K-means clustering
•  Spectral clustering
–  Density Estimation
•  Gaussian mixture model (GMM)
•  Graphical models
–  Dimensionality reduction
•  Principal component analysis (PCA)
•  Factor analysis

https://www.ibm.com/developerworks/library/cc-models-machine-learning/index.html
Reinforcement learning
The Value Function: defines how
to choose a “good” action.

https://www.ibm.com/developerworks/library/cc-models-machine-learning/index.html
Summary

Deep Learning
27

Deep Learning (Deep Neural Networks)
“Deep Learning” means using a neural network
with several layers of nodes between input and output
Why is it generally better than other methods on image, speech and certain
other types of data?
The series of layers between input & output do feature identification and
processing in a series of stages, just as our brains seem to.
https://www.slideshare.net/LuMa921/deep-learning-a-visual-introduction

Deep Learning
Silver, David, et al. "Mastering the game of Go with deep neural networks and tree search." nature 529.7587 (2016): 484-489.

Deep Learning software
1.  Theano – CPU/GPU symbolic expression compiler in python (from MILA lab at
University of Montreal)
2.  Torch – provides a Matlab-like environment for state-of-the-art machine learning
algorithms in lua (from Ronan Collobert, Clement Farabet and Koray Kavukcuoglu)
3. Tensorflow – TensorFlow™ is an open source software library for numerical
computation using data flow graphs.
4.  Caffe -Caffe is a deep learning framework made with expression, speed, and
modularity in mind.
5.  Keras– A theano based deep learning library.
6.  Chainer – A GPU based Neural Network Framework
7.  Matlab Deep Learning – Matlab Deep Learning Tools
8.  CNTK – Computational Network Toolkit – is a unified deep-learning toolkit by
Microsoft Research.
9.  MatConvNet – A MATLAB toolbox implementing Convolutional Neural
Networks (CNNs) for computer vision applications. It is simple, efficient, and can
run and learn state-of-the-art CNNs.
10.  DeepLearnToolbox – A Matlab toolbox for Deep Learning (from Rasmus Berg
Palm)
http://deeplearning.net/software_links/

Thank you
Happy Learning!
Lecturer: Dr Xiaonan Wang
Assistant Professor
Department of Chemical and Biomolecular Engineering
National University of Singapore
NUS Faculty of Engineering, Block E5, Unit #03-04
+65 6601 6221 (Tel) chewxia@nus.edu.sg
32

Lecture on AI and Machine Learning

More Related Content

What's hot

Similar to Lecture on AI and Machine Learning

Recently uploaded

Lecture on AI and Machine Learning