Linear Regression

Objectives
v What is machine learning
vTypes of data and terminology
vTypes of machine learning
v Supervised learning
v Linear regression
v Least square and Gradient Descent
v Hands on implementing Linear Regression from sketch.

Machine Learning
v Machine Learning is the science to make computers learn from data
without explicitly program them and improve their learning over time
in autonomous fashion.
v This learning comes by feeding them data in the form of
observations and real-world interactions.”
v Machine Learning can also be defined as a tool to predict future
events or values using past data.

Types of Data
vBased on Values
vContinuous data (ex. Age – 0-100)
v Categorical data (ex. Gender- Male/Female)
v Based on pattern
v Structured data (ex. Databases)
v Unstructured data (ex. Audio, Video, Text)

Types of Data- continued
v Labelled data – consists of input output pair. For every set
input features the output/response/label is present in
dataset. (ex- labelled image as cat’s or dog’s photo)
{ 𝑥#, 𝑦# , 𝑥&, 𝑦& , 𝑥', 𝑦' … … … … … 𝑥), 𝑦) }
vUnlabelled data- There is no output/response/label for the
input features in data. (ex. news articles, tweets, audio)
{𝑥#, 𝑥&, 𝑥' … … … … 𝑥)}

Types of Data- continued
vTraining Data – Sample data points which are used to train the
machine learning model.
vTest Data- sample data points that are used to test the
performance of machine learning model.
Note- For modelling, the original dataset is partitioned into the ratio of 70:30 or 75:25 as
training data and test data.

Types of Machine Learning
Machine
Learning
Supervised
Learning
Regression Classification
Unsupervised
Learning
Clustering PCA
Reinforcement
Learning

Supervised Learning
v Class of machine learning that work on externally supplied instances in form of
predictor attributes and associated target values.
v The model learns from the training data using these ‘target variables’ as
reference variables.
vEx1 : model to predict the resale value of a car based on its mileage, age,
color etc.
v The target values are the ‘correct answers’ for the predictor model which can
either be a regression model or a classification model.

Motivation for learning
v It is being assumed that there exists a relationship/association between input
features and target variable.
v Relationship can be observed by plotting a scatter plot between the two
variables.
v Relationship measure can be quantified by calculating correlation between two
the variables.
𝑐𝑜𝑟𝑟 𝑥, 𝑦 =
𝑐𝑜𝑣(𝑥, 𝑦)
𝑣𝑎𝑟 𝑥 . 𝑣𝑎𝑟(𝑦)
=
∑ 𝑥5 − ̅
𝑥 ∗ 𝑦5 − 9
𝑦
∑ 𝑥5 − ̅
𝑥
&
∑ 𝑦5 − 9
𝑦
&

Linear Regression
v Linear regression is a way to identify a relationship between two or more
variables and use these relationships to predict values of one variable for given
value(s) of other variable(s).
v Linear regression assume the relationship between variables can be modelled
through linear equation or an equation of line
𝒚 = 𝒘𝟎 + 𝒘𝟏𝑿 Independent/Regressor variable
Dependent/Regressed variable
Intercept
Slope

Multiple Regression
v Last slide showed the linear regression model with one independent and one
dependent variable.
v In Real world a data point has various important attributes and they need to be
catered to while developing a regression model. (Many independent variables and
one dependent variable)
𝑦 = 𝑤A + 𝑤#𝑥# + 𝑤&𝑥& + 𝑤'𝑥'. … … … . wCxC

Regression –Problem Formulation
Let you have given with a data:
110
120
130
140
150
160
170
35 45 55 65 75
Blood
Pressure
(Y)
Age in Year (X)
Age in Years
(X)
Blood Pressure
(Y)
56 147
49 145
72 160
38 115
63 130
47 128

Linear Regression
v For given example the Linear Regression is modeled as:
𝐵𝑙𝑜𝑜𝑑𝑃𝑟𝑒𝑠𝑠𝑢𝑟𝑒 𝑦 = 𝑤A + 𝑤#𝐴𝑔𝑒𝑖𝑛𝑌𝑒𝑎𝑟(𝑋)
OR
𝑦 = 𝑤A + 𝑤#𝑋 – Equation of line
with 𝑤A 𝑖𝑠 𝑖𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 𝑜𝑛 𝑌_𝑎𝑥𝑖𝑠 𝑎𝑛𝑑 𝑤# 𝑖𝑠 𝑠𝑙𝑜𝑝𝑒 𝑜𝑓 𝑙𝑖𝑛𝑒
Blood Pressure - Dependent Variable
Age in Year - Independent Variable

Linear Regression- Best Fit Line
v Regression uses line to show the trend of distribution.
vThere can be many lines that try to fit the data points in scatter diagram
v The aim is to find Best fit Line
110
120
130
140
150
160
170
35 45 55 65 75
Blood
Pressure
(Y)
Age in Year (X)

What is Best Fit Line
v Best fit line tries to explain the variance in given data. (minimize the total residual/error)

Linear Regression- Methods to Get Best
vLeast Square
v Gradient Descent

Linear Regression- Least Square
Model: 𝑌 = 𝑤A + 𝑤#𝑋
Task: Estimate 𝑡ℎ𝑒 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑤A 𝑎𝑛𝑑 𝑤#
According to pr𝑖𝑛𝑐𝑖𝑝𝑙𝑒 𝑜𝑓 𝑙𝑒𝑎𝑠𝑡 𝑠𝑞𝑢𝑎𝑟𝑒 𝑡ℎ𝑒 𝑛𝑜𝑟𝑚𝑎𝑙 𝑒𝑞𝑢𝑎𝑡𝑖𝑜𝑛𝑠 𝑡𝑜 𝑠𝑜𝑙𝑣𝑒 𝑓𝑜𝑟 𝑤A 𝑎𝑛𝑑 𝑤#
c
5d#
)
𝑌5 = 𝑛 𝑤A + 𝑤# c
5d#
)
𝑋5 … … … … … (1)
c
5d#
)
𝑋5𝑌5 = 𝑤A c
5d#
)
𝑋5 + 𝑤# c
5d#
)
𝑋5
&
… … … … . (2)

Linear Regression–Least Square
Let divide the equation (1) by n (number of sample points) we get:
1
𝑛
c
5d#
)
𝑌5 = 𝑤A + 𝑤#
1
𝑛
c
5d#
)
𝑋5
OR
9
𝑦 = 𝑤A + 𝑤# ̅
𝑥………….(3)
So line of regression will always passes through the points ( ̅
𝑥, 9
𝑦)

Now we know :
𝑐𝑜𝑣 𝑥, 𝑦 =
#
)
∑5d#
)
𝑥5𝑦5 − ̅
𝑥9
𝑦 ≔
#
)
∑5d#
)
𝑥5𝑦5 = 𝑐𝑜𝑣 𝑥, 𝑦 + ̅
𝑥9
𝑦 ………(4)
and
𝑣𝑎𝑟 𝑥 =
#
)
∑5d#
)
𝑥5
&
− ̅
𝑥& and 𝑣𝑎𝑟 𝑦 =
#
)
∑5d#
)
𝑦5
&
− 9
𝑦&
Dividing equation (2) by n and using equation (4) and (5) we get:
𝑐𝑜𝑣 𝑥, 𝑦 + ̅
𝑥9
𝑦 = 𝑤A ̅
𝑥 + 𝑤#(𝑣𝑎𝑟 𝑥 + ̅
𝑥&)…………………….(5)

Now by using equation
9
𝑦 = 𝑤A + 𝑤# ̅
𝑥
and
𝑐𝑜𝑣 𝑥, 𝑦 + ̅
𝑥9
𝑦 = 𝑤A ̅
𝑥 + 𝑤#(𝑣𝑎𝑟 𝑥 + ̅
𝑥&)
We will get:
𝑤# =
𝑐𝑜𝑣(𝑥, 𝑦)
𝑣𝑎𝑟(𝑥)
and 𝑤A = 9
𝑦 − 𝑤# ̅
𝑥

Performance metric for least square regression
𝑅klm
&
= 1 −
(1 − 𝑅&)(𝑛 − 1)
(𝑛 − 𝑘 − 1)
𝑅& = 1 −
1
𝑛
∑5d#
)
(𝑦5 − 𝑦ℎ𝑎𝑡5)&
1
𝑛
∑5d#
) (𝑦5 − 9
𝑦)&

Linear Regression- Gradient Descent
Define the cost function,
𝑐𝑜𝑠𝑡 𝑤A, 𝑤# =
1
𝑛
c
5d#
)
(𝑦5 − 𝑦ℎ𝑎𝑡5)&
Objective of gradient Descent
𝐦𝐢𝐧
𝒘𝟎,𝒘𝟏
1
𝑛
c
5d#
)
(𝑦5 − (𝑤A + 𝑤#𝑥5))&

the objective,
𝐦𝐢𝐧
𝒘𝟎,𝒘𝟏
1
𝑛
c
5d#
)
(𝑦5 − (𝑤A + 𝑤#𝑥5))&
Cost(w0,
w1)
w0

v Gradient descent works if following steps:
1. Initialize the parameters to some random variable
2. Calculate the gradient of cost function w. r. t. to parameters
3. Update the parameters using gradient in opposite direction.
4. Repeat step-2 and step-3 for some number of times or till it reaches to minimum
cost value.

1
𝑛
c
5d#
)
(𝑦5 − (𝑤A + 𝑤#𝑥5))&
Calculating gradients of cost function:
𝑔𝑟𝑎𝑑𝑤A =
𝜕𝑐𝑜𝑠𝑡(𝑤A, 𝑤#)
𝜕𝑤A
=
2
𝑛
c
5d#
)
(𝑦5 − (𝑤A + 𝑤#𝑥5))(−1)
𝑔𝑟𝑎𝑑𝑤# =
𝜕𝑐𝑜𝑠𝑡(𝑤A, 𝑤#)
𝜕𝑤#
=
2
𝑛
c
5d#
)
(𝑦5 − (𝑤A + 𝑤#𝑥5))(−𝑥)
Parameter update:
𝑤A = 𝑤A − 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑟𝑎𝑡𝑒 ∗ 𝑔𝑟𝑎𝑑𝑤A
𝑤# = 𝑤# − 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑟𝑎𝑡𝑒 ∗ 𝑔𝑟𝑎𝑑𝑤#

Performance metric for gradient based regression
𝑅𝑀𝑆𝐸 =
(𝑦5 − 𝑦ℎ𝑎𝑡5)&
𝑛
Mean absolute error (MAE) is a measure of difference between two variables.
𝑀𝐴𝐸 =
𝑦5 − 𝑦ℎ𝑎𝑡5
𝑛
Root Mean Square Error (RMSE) is the standard deviation of prediction errors.

Thank you !
Let see the hands on…

Regression
(Parametric Data Reduction: Regression and Log-Linear
Models)
Linear regression
• Data modeled to fit a straight line
• Often uses the least-square method to fit the line
Multiple regression
• Allows a response variable Y to be modeled as a linear
function of multidimensional feature vector
Log-linear model
• Approximates discrete multidimensional probability
distributions

Regress Analysis and Log-Linear Models
• Linear regression: Y = w X + b
• Two regression coefficients, w and b, specify the line and are to be estimated by using
the data at hand
• Using the least squares criterion to the known values of Y1, Y2, …, X1, X2, ….
• Multiple regression: Y = b0 + b1 X1 + b2 X2
• Many nonlinear functions can be transformed into the above
• Log-linear models:
• The technique is used for both hypothesis testing and model building.
• Log-linear analysis is a technique used in statistics to examine the relationship between
more than two categorical variables.
• Log-linear analysis is a multidimensional extension of the classical cross-tabulation chi-
square test .
• Approximate discrete multidimensional probability distributions
• Estimate the probability of each point (tuple) in a multi-dimensional space for a set of
discretized attributes, based on a smaller subset of dimensional combinations

Linear Algebra
• Linear algebra is the branch of mathematics concerning vector spaces and
linear mappings between such spaces.
 Why we do study linear algebra?
• Provides a way to compactly represent and operate on sets of linear
equations.
• In machine learning, we represents data as a matrices.
 Consider the following system of the equations:
4X1-5X2= -13
-2X1+3X2=9
In Matrix Notation ,

Applications of Linear Algebra
• Matrices in Engineering, such as a line of springs.
• Graphs and Networks, such as analyzing networks.
• Markov Matrices, Population, and Economics, such as population growth.
• Linear Programming, the simplex optimization method.
• Fourier Series: Linear Algebra for functions, used widely in signal
processing.
• Linear Algebra for statistics and probability, such as least squares for
regression.
• Computer Graphics, such as the various translation, rescaling and rotation
of images.
• Application of linear algebra is that it is the type of mathematics used by
Albert Einstein in parts of his theory of relativity.

Examples of Linear algebra in Lachine Learning
• Dataset and Data Files
• Images and Photographs
• One Hot Encoding
• Linear Regression
• Regularization
• Principal Component Analysis (PCA)
• Singular-Value Decomposition (SVD)
• Latent Semantic Analysis
• Recommender Systems

Mathematical Optimization
• Mathematical optimization is the selection of a best element from set of available
alternatives.
Represented as: min f0(x) , x is optimization variable, and f0 is objective function.
f0(x) is the minimum possible values in the feasible region.
Example:
• Data Fitting:
Variable: Parameter of the model.
Constraint: Parameter Limits, Prior information
Objective: Measures of fit(E.g. Minimizing of error)

Solving Optimization Problem
• Optimization is the very tough problems to solve.
• Optimization problems are classified in to various classes based on the
properties of objectives and constraints.
• Some of these classes can be solved efficiently :
1)Linear program
II) Least Square problem
III) Convex Optimization Problem

Data Visualization
• “Seeing” the Data

Visual Presentation
• For any kind of high dimensional data set, displaying predictive
relationships is a challenge.
• The picture on the previous slide uses 3-D graphics to portray the
weather balloon data numbers in text Table 11-4. We learn very little
from just examining the numbers .
• Shading is used to represent relative degrees of thunderstorm
activity, with the darkest regions the heaviest activity.

Market Basket Analysis
• This is the most widely used machine learning techniques.
• It essentially determines what products people purchase together.
• Stores can use this information to place these products in the same area.
• Direct marketers can use this information to determine which new products
to offer to their current customers.
• Inventory policies can be improved if reorder points reflect the demand for
the complementary products.
We first need a list of transactions and what was purchased. This is pretty
easily obtained these days from scanning cash registers.
Next, we choose a list of products to analyze, and tabulate how many times
each was purchased with the others.
The diagonals of the table shows how often a product is purchased in any
combination, and the off-diagonals show which combinations were bought.

A Convenience Store Example
(5 transactions)
Consider the following simple example about five transactions at a
convenience store:
Transaction 1: Frozen pizza, cola, milk
Transaction 2: Milk, potato chips
Transaction 3: Cola, frozen pizza
Transaction 4: Milk, pretzels
Transaction 5: Cola, pretzels
These need to be cross tabulated and displayed in a table.

A Convenience Store Example (5 transactions)

Using the Results
• The tabulations can immediately be translated into association rules
and the numerical measures computed.
• Comparing this week’s table to last week’s table can immediately
show the effect of this week’s promotional activities.
• Some rules are going to be trivial (hot dogs and buns sell together) or
inexplicable (toilet rings sell only when a new hardware store is
opened).
• Limitations to Market Basket Analysis:
• A large number of real transactions are needed to do an effective
basket analysis, but the data’s accuracy is compromised if all the
products do not occur with similar frequency.
• The analysis can sometimes capture results that were due to the
success of previous marketing campaigns (and not natural tendencies
of customers).

Computing Measures of Association

Data Preprocessing
• Data quality: accuracy, completeness, consistency, timeliness, believability,
interpretability
• Data cleaning: e.g. missing/noisy values, outliers
• Data integration from multiple sources:
• Entity identification problem
• Remove redundancies
• Detect inconsistencies
• Data reduction
• Dimensionality reduction
• Numerosity reduction
• Data compression
• Data transformation and data discretization
• Normalization
• Concept hierarchy

Why Preprocess the Data?
• Measures for data quality: A multidimensional view
• Accuracy: correct or wrong, accurate or not
• Completeness: not recorded, unavailable, …
• Consistency: some modified but some not, dangling, …
• Timeliness: timely update?
• Believability: how trustable the data are correct?
• Interpretability: how easily the data can be understood?

Data Transformation
• A function that maps the entire set of values of a given attribute to a new set of replacement values s.t.
each old value can be identified with one of the new values
• Methods
• Smoothing: Remove noise from data
• Attribute/feature construction
• New attributes constructed from the given ones
• Aggregation: Summarization, data cube construction
• Normalization: Scaled to fall within a smaller, specified range
• min-max normalization
• z-score normalization
• normalization by decimal scaling
• Discretization: Concept hierarchy climbing

Discretization
• Three types of attributes
• Nominal—values from an unordered set, e.g., color, profession
• Ordinal—values from an ordered set, e.g., military or academic rank
• Numeric—real numbers, e.g., integer or real numbers
• Discretization: Divide the range of a continuous attribute into intervals
• Interval labels can then be used to replace actual data values
• Reduce data size by discretization
• Supervised vs. unsupervised
• Split (top-down) vs. merge (bottom-up)
• Discretization can be performed recursively on an attribute
• Prepare for further analysis, e.g., classification

Data Augmentation
• Data augmentation is a data-depended process.
• In general, you need it when your training data is complex and you have a
few samples.
• It can help, the features the network will learn to extract are easy and
highly different from each other.
• It means increasing the number of data points.
• In terms of images, it may mean that increasing the number of images in the
dataset.
• It adds value to base data by adding information derived from internal and
external sources.
• Crops random regions of the source image according to a given range and
speed, to generate a smooth animation.

Linearity vs non Linearity
Linearity:
Linearity and non linearity defined on the basis of activation function.
• In linear regression, data are modeled using linear predictor functions.
• linear methods involve only linear combinations of data, leading to easier implementations,
etc.
• it is a statistical analytic technique used for separating sets of observed values and allocating
new values.
• It is used PCA and creating an Index for missing data.
• Non Linearity: ( why do we need Non-Linearities)
• Non-linear functions are those which have degree more than one and they have a curvature.
• when we plot a Non-Linear function. Now we need a Neural Network Model to learn and
represent almost anything and any arbitrary complex function which maps inputs to outputs.
• Non linear techniques are ANN and SVM.
• They use, in general, non linear functions of the data (the activation function in ANN or the
kernel in SVM .

General discussion
• Pattern recognition
• Patterns: images, personal records, driving habits, etc.
• Represented as a vector of features (encoded as integers or real
numbers in NN)
• Pattern classification:
• Classify a pattern to one of the given classes
• Form pattern classes
• Pattern associative recall
• Using a pattern to recall a related pattern
• Pattern completion: using a partial pattern to recall the whole
pattern
• Pattern recovery: deals with noise, distortion, missing
information

• If n = 3 (three input units), then the decision boundary is a two
dimensional plane in a three dimensional space
• In general, a decision boundary is a n-1
• dimensional hyper-plane in an n dimensional space, which partition
the space into two decision regions
• This simple network thus can classify a given pattern into one of the
two classes, provided one of these two classes is entirely in one
decision region (one side of the decision boundary) and the other
class is in another region.
• The decision boundary is determined completely by the weights W
and the bias b (or threshold q).
0
1

  
n
i i
i w
x
b

Linear Separability Problem
• If two classes of patterns can be separated by a decision boundary,
represented by the linear equation
then they are said to be linearly separable. The simple network can
correctly classify any patterns.
• Decision boundary (i.e., W, b or q) of linearly separable classes can be
determined either by some learning procedures or by solving linear
equation systems based on representative patterns of each classes.
• If such a decision boundary does not exist, then the two classes are said
to be linearly inseparable.
• Linearly inseparable problems cannot be solved by the simple network ,
more sophisticated architecture is needed.
0
1

  
n
i i
i w
x
b

Activation Function
• What are Activation functions and what are it uses in a Neural Network Model?
• Activation functions are really important for a Artificial Neural Network to learn
and make sense of something really complicated and Non-linear complex functional
mappings between the inputs and response variable.
• It is also known as Transfer Function.
• Activation functions are important for a Artificial Neural Network to learn and
understand the complex patterns.
• The main function of it is to introduce non-linear properties into the network.
• What it does?
• Their main purpose is to convert a input signal of a node in a A-NN to an output
signal. That output signal now is used as a input in the next layer in the stack.
• The non linear activation function will help the model to understand the complexity
and give accurate results.

• why can’t we do it without activating the input signal?
• If we do not apply a Activation function then the output signal would
simply be a simple linear function.
• A linear function is just a polynomial of one degree. Now, a linear
equation is easy to solve but they are limited in their complexity and have
less power to learn complex functional mappings from data.
• A Neural Network without Activation function would simply be a Linear
regression Model, which has limited power and does not performs good
most of the times.
• Also without activation function our Neural network would not be able to
learn and model other complicated kinds of data such as images, videos ,
audio , speech etc.

3-Types of Activation Functions
Binary Step Function
• A binary step function is a threshold-based activation function. If the
input value is above or below a certain threshold, the neuron is
activated and sends exactly the same signal to the next layer.

• Linear Activation Function
A linear activation function takes the form: A = cx

Non-Linear Activation Functions
• Non-linear functions are those which have degree more than one and they have a curvature.
• when we plot a Non-Linear function. Now we need a Neural Network Model to learn and
represent almost anything and any arbitrary complex function which maps inputs to outputs.
• Non linear techniques are ANN and SVM.
• Modern neural network models use non-linear activation functions.
• They allow the model to create complex mappings between the network’s inputs and outputs,
which are essential for learning and modeling complex data, such as images, video, audio, and data
sets which are non-linear or have high dimensionality.
• Almost any process imaginable can be represented as a functional computation in a neural
network, provided that the activation function is non-linear.
• Non-linear functions address the problems of a linear activation function: They allow
backpropagation because they have a derivative function which is related to the inputs.
• They allow “stacking” of multiple layers of neurons to create a deep neural network. Multiple
hidden layers of neurons are needed to learn complex data sets with high levels of accuracy.
• Most popular types of Activation functions -
• Sigmoid or Logistic
• Hyperbolic tangent
• ReLu -Rectified linear units

Sigmoid Activation function
• It is a activation function of form f(x) = 1 / 1 + exp(-x) .
• Its Range is between 0 and 1. It is a S — shaped curve.
• It is easy to understand and apply but it has major reasons which have made it fall out
of popularity -
• Vanishing gradient problem
• Secondly , its output isn’t zero centered. It makes the gradient updates go too far in
different directions. 0 < output < 1, and it makes optimization harder.
• Sigmoids saturate and kill gradients.
• Sigmoids have slow convergence.

Sigmoid / Logistic
Advantages
• Smooth gradient, preventing “jumps” in output values.
• Output values bound between 0 and 1, normalizing the output of each
neuron.
• Clear predictions—For X above 2 or below -2, tends to bring the Y value
(the prediction) to the edge of the curve, very close to 1 or 0. This enables
clear predictions.
Disadvantages
• Vanishing gradient—for very high or very low values of X, there is almost
no change to the prediction, causing a vanishing gradient problem. This can
result in the network refusing to learn further, or being too slow to reach
an accurate prediction.
• Outputs not zero centered.
• Computationally expensive

Hyperbolic Tangent function- Tanh
It’s mathamatical formula is f(x) = 1 — exp(-2x) / 1 + exp(-2x).
• Now it’s output is zero centered because its range in between -1 to 1
• ie -1 < output < 1 .
• Hence optimization is easier in this method hence in practice it is always preferred over
Sigmoid function . But still it suffers from Vanishing gradient problem.

ReLu- Rectified Linear units
• It has become very popular due to 6 times improvement in convergence from Tanh
function.
• It’s just R(x) = max(0,x) i.e if x < 0 , R(x) = 0 and if x >= 0 , R(x) = x.
• Hence as seeing the mathematical form of this function we can see that it is very simple
and efficient .
• A lot of times in Machine learning and computer science we notice that most simple and
consistent techniques and methods are only preferred and are best.
• Hence it avoids and rectifies vanishing gradient problem . Almost all deep learning
Models use ReLu nowadays.
• But its limitation is that it should only be used within Hidden layers of a Neural
Network Model.
• For output layers we should use a Softmax function for a Classification
problem to compute the probabilities for the classes.
• For a regression problem it should simply use a linear function.

• Another problem with ReLu is that some gradients can be breakable during training
and can die.
• It can cause a weight update which will makes it never activate on any data point
again. Simply saying that ReLu could result in Dead Neurons.
• To fix this problem another modification was introduced called Leaky ReLu to fix the
problem of dying neurons. It introduces a small slope to keep the updates alive.
• We then have another variant made form both ReLu and Leaky ReLu
called Maxout function .

Neural networks
• Neural networks are made up of many artificial neurons.
• Each input into the neuron has its own weight associated with
it illustrated by the red circle.
• A weight is simply a floating point number and it's these we
adjust when we eventually come to train the network.
2

Neural networks
• A neuron can have any number of inputs from one to
n, where n is the total number of inputs.
• The inputs may be represented therefore as x1, x2, x3…
xn.
• And the corresponding weights for the inputs as w1,
w2, w3… wn.
• Output a = x1w1+x2w2+x3w3... +xnwn
3

How do we actually use an artificial
neuron?
• feedforward network: The neurons in each layer feed their
output forward to the next layer until we get the final output
from the neural network.
• There can be any number of hidden layers within a
feedforward network.
• The number of neurons can be completely arbitrary.
4

Multi-Layer Perceptron (MLP)
5

We will introduce the MLP and the backpropagation
algorithm which is used to train it
MLP used to describe any general feedforward (no
recurrent connections) network
However, we will concentrate on nets with units
arranged in layers
x1
xn
6

NB different books refer to the above as either 4 layer (no.
of layers of neurons) or 3 layer (no. of layers of adaptive
weights). We will follow the latter convention
1st question:
what do the extra layers gain you? Start with looking at
what a single layer can’t do
x1
xn
7

Perceptron Learning Theorem
• A perceptron (threshold unit) can learn
anything that it can represent (i.e. anything
separable with a hyperplane)
8

The Exclusive OR problem
A Perceptron cannot represent Exclusive OR
since it is not linearly separable.
9

Minsky & Papert (1969) offered solution to XOR problem by
combining perceptron unit responses using a second layer of
Units. Piecewise linear classification using an MLP with
threshold (perceptron) units
1
2
+1
+1
3
10

xn
x1
x2
Input
Output
Three-layer networks
Hidden layers 11

Properties of MLP architecture
• No connections within a layer
• No direct connections between input and output layers
• Fully connected between layers
• Often more than 3 layers
• Number of output units need not equal number of input units
• Number of hidden units per layer can be more or less than
input or output units
y f w x b
i ij j i
j
m
 


( )
1
Each unit is a perceptron
Often include bias as an extra weight 12

What do each of the layers do?
1st layer draws
linear boundaries
2nd layer combines
the boundaries
3rd layer can generate
arbitrarily complex
boundaries 13

Backward pass phase: computes ‘error signal’, propagates
the error backwards through network starting at output units
(where the error is the difference between actual and desired
output values)
Forward pass phase: computes ‘functional signal’, feed forward
propagation of input pattern signals through network
Backpropagation learning algorithm ‘BP’
Solution to credit assignment problem in MLP. Rumelhart, Hinton and
Williams (1986) (though actually invented earlier in a PhD thesis
relating to economics)
BP has two phases:
14

Conceptually: Forward Activity -
Backward Error
15

Forward Propagation of Activity
• Step 1: Initialise weights at random, choose a
learning rate η
• Until network is trained:
• For each training example i.e. input pattern and
target output(s):
• Step 2: Do forward pass through net (with fixed
weights) to produce output(s)
– i.e., in Forward Direction, layer by layer:
• Inputs applied
• Multiplied by weights
• Summed
• ‘Squashed’ by sigmoid activation function
• Output passed to each neuron in next layer
– Repeat above until network output(s) produced
16

Step 3. Back-propagation of error
• Compute error (delta or local gradient) for each
output unit δ k
• Layer-by-layer, compute error (delta or local
gradient) for each hidden unit δ j by backpropagating
errors (as shown previously)
Step 4: Next, update all the weights Δwij
By gradient descent, and go back to Step 2
 The overall MLP learning algorithm, involving
forward pass and backpropagation of error
(until the network training completion), is
known as the Generalised Delta Rule (GDR),
or more commonly, the Back Propagation
(BP) algorithm
17

‘Back-prop’ algorithm summary
(with Maths!) (Not Examinable)
18

‘Back-prop’ algorithm summary
(with NO Maths!)
19

Worked example: Forward Pass
21

Worked example: Forward Pass
22

Worked example: Backward Pass
23

Worked example: Update Weights
Using Generalized Delta Rule (BP)
24

Similarly for the all weights wij:
25

Training
• This was a single iteration of back-prop
• Training requires many iterations with many
training examples or epochs (one epoch is
entire presentation of complete training set)
• It can be slow !
• Note that computation in MLP is local (with
respect to each neuron)
• Parallel computation implementation is also
possible
27

Training and testing data
• How many examples ?
– The more the merrier !
• Disjoint training and testing data sets
– learn from training data but evaluate
performance (generalization ability) on
unseen test data
• Aim: minimize error on test data
28

weights and bias
• Bias is just like an intercept added in a linear equation. It is an additional parameter in the Neural
Network which is used to adjust the output along with the weighted sum of the inputs to the neuron.
• bias value allows you to shift the activation function to either right or left.
output = sum (weights * inputs) + bias
• The output is calculated by multiplying the inputs with their weights and then passing it through an
activation function like the Sigmoid function, etc. Here, bias acts like a constant which helps the model to
fit the given data. The steepness of the Sigmoid depends on the weight of the inputs.
It allows you to move the line down and up fitting the prediction with the data better.
• If bias=0 ie , the constant c is absent then the line will pass through the origin (0, 0) and you will get a
poorer fit. (Neural Network is an important component of Machine Learning)

• If the inputs are x1,x2,x3, then the weight applied to them are denoted
as w1,w2,w3. Then the output is as follows: y = f(x) = Σxiwi
• Bias is like the intercept added in a linear equation. It is an additional
parameter in the Neural Network which is used to adjust the output
along with the weighted sum of the inputs to the neuron. Thus, Bias is a
constant which helps the model in a way that it can fit best for the given
data.
• The processing done by the neuron is:
output = sum (weights * inputs) + bias
For example, consider an equation y=mx+c
Here m is acting as weight and the constant c is acting as bias.
• Bias units are not connected to any previous layer, it is just appended to
the start/end of the input and each hidden layer, and is not affected by
the values in the previous layer. Explain with diagram:

Here x1,x2,x3 are inputs and w1, w2,w3 are weights. It takes an input, processes it, passes it through an
activation function, and returns the output.

Loss function
• If predictions deviates too much from actual results.
• It improve with the help of some optimization function, loss function learns to reduce
the error in prediction.
Type of loss functions: It can be classified into two major categories.
Regression losses and Classification losses.
Classification: we are trying to predict output from set of finite categorical values i.e
Given large data set of images of hand written digits, categorizing them into one of 0–9
digits.
Regression, on the other hand, deals with predicting a continuous value for example
given floor area, number of rooms, size of rooms, predict the price of room.
n - Number of training examples.
i - ith training example in a data set.
y(i) - Ground truth label for ith training example.
y_hat(i) - Prediction for ith training example.

• Regression Losses:
• Mean Square Error/Quadratic Loss/L2 Loss
• Mathematical formulation :-
• Mean square error is measured as the average of squared difference between
predictions and actual observations.
• Due to squaring, predictions which are far away from actual values are penalized
heavily in comparison to less deviated predictions.
• MSE has nice mathematical properties which makes it easier to calculate gradients.

• Mean Absolute Error/L1 Loss
• Mathematical formulation :-
• Mean absolute error: It is measured as the average of sum of absolute differences
between predictions and actual observations.
• This as well measures the magnitude of error without considering their direction.
• MAE needs more complicated tools such as linear programming to compute the gradients.
(MAE is more robust to outliers since it does not make use of square)
• Mean Bias Error: This is same as MSE, with the only difference that we don’t take
absolute values. it could determine if the model has positive bias or negative bias.

• Classification Losses: (Multi class SVM Loss)
• It is used for maximum-margin classification, most notably for support vector machines.
Although not differentiable, it’s a convex function which makes it easy to work with usual
convex optimizers used in machine learning domain.
Consider an example where we have three training examples and three classes to predict — Dog, cat
and horse. Below the values predicted by our algorithm for each of the classes :

Gradient Descent Algorithm (GDA)
• Gradient descent is one of the most popular algorithms to perform optimization and by
far the most common way to optimize neural networks. It is an iterative optimization
algorithm used to find the minimum value for a function.
Consider that you are walking along the graph below, and you are currently at the ‘green’
dot. Your aim is to reach the minimum i.e the ‘red’ dot, but from your position, you are
unable to view it. Why we need of GDA? (To reduce cost function of higher degree.)

Machine Learning Model
• Consider a bunch of data points in a 2 D space. Assume that the data is related to the height and weight of a
group of students.
• We are trying to predict some kind of relationship between these quantities so that we could predict the weight
of some new students afterwards.
This is simple example of a supervised Machine Learning technique.
Exam: Let us now draw an arbitrary line in space that passes through some of these data points. The equation of
this straight line would be Y = mX + b where m is the slope and b is its intercept on the Y-axis.

Predictions
• Given a known set of inputs and their corresponding outputs, A machine learning model
tries to make some predictions for a new set of inputs.
This relates to the a Cost function or Loss function.

Cost Function
• A Cost Function/Loss Function evaluates the performance of our Machine Learning Algorithm.
• The Loss function computes the error for a single training example while the Cost function is the average of
the loss functions for all the training examples.
(A Cost function basically tells us ‘ how good’ our model is at making predictions for a given value of m and b.)
Lets ,there are a total of ’N’ points in the dataset and for all those ’N’ data points we want to
minimize the error. So the Cost function would be the total squared error i.e
Why do we take the squared differences and simply not the absolute differences?
Because the squared differences make it easier to derive a regression line.
Indeed, to find that line we need to compute the first derivative of the Cost function,
and it is much harder to compute the derivative of absolute values than squared
values. Also, the squared differences increase the error distance, thus, making the
bad predictions more pronounced than the good ones.

Minimizing the Cost Function
• The goal of any Machine Learning Algorithm is to minimize the Cost Function.
• This is because a lower error between the actual and the predicted values signifies that
the algorithm has done a good job in learning. Since we want the lowest error value, we
want those‘ m’ and ‘b’ values which give the smallest possible error.

• To minimize the function above, we need to find that value of X that produces the
lowest value of Y.
• which is the red dot. It is quite easy to locate the minima here since it is a 2D graph but
this may not always be the case especially in case of higher dimensions. For those
cases, we need to devise an algorithm to locate the minima, and that algorithm is
called Gradient Desce
• Gradient descent.
.Consider that you are walking along the graph below, and you are currently at the ‘green’
dot. Your aim is to reach the minimum i.e the ‘red’ dot, but from your position, you are
unable to view it.

• Possible actions would be:
• You might go upward or downward
• If you decide on which way to go, you might take a bigger step or a little step to reach
your destination.
• Gradient Descent Algorithm helps us to make these decisions efficiently and effectively
with the use of derivatives.
• A derivative is a term that comes from calculus and is calculated as the slope of the graph
at a particular point.
• The slope is described by drawing a tangent line to the graph at the point.
• So, if we are able to compute this tangent line, we might be able to compute the desired
direction to reach the minima.

• The Minimum Value
• In the same figure, if we draw a tangent at the green point, we know that
if we are moving upwards, we are moving away from the minima and vice
versa. Also, the tangent gives us a sense of the steepness of the slope.
The slope at the blue point is less steep than that at the green point which means it will take much
smaller steps to reach the minimum from the blue point than from the green point.

• Mathematical Interpretation of Cost Function

Auto-encoders
• Neural network to reconstruct the input
• An auto encoder is a type of artificial neural network used to
learn efficient data coding in an unsupervised manner.
• The objective of an auto encoder is to learn a representation
(encoding) for a set of data, typically for dimensionality
reduction, by training the network.
Use of Auto encoder:
• Features reduction
• Unsupervised pre-training
• Problem of inter-subject variation

• Neural network to reconstruct the input
• Seems easy , no ??? Well, yes, but…
• Constraining the network yields interesting results

auto-encoders
• Suppose we have a set of
unlabeled training examples
• We want to train the
network to output an
estimation of the input
ො
𝑥 = ℎ𝑊,𝑏 ≈ 𝑥
𝑥1, 𝑥2, 𝑥3, …

Auto-encoders – examples
• Web images search

Auto-encoders – examples
•Anomaly detection
•Train on true images
and find outliers by
error.
Error

Auto-encoders – how it works?
1. train a sparse autoencoder
on the raw inputs
to find primary features
𝑥𝑘
ℎ𝑘
(1)

2. use primary features as the
"raw input" to another sparse
autoencoder to learn secondary
features
(next layer is smaller than the
current layer)
2a. Repeat for N layers (if needed)

3. Input last layer features into
a softmax classifier

1. Combine all layers

Emotion Recognition BCI
• Brain Computer Interface - BCI
• Better communication with computers for disable people.
• Detect patients' emotional responses to specific inexpressive
faces.

Emotion Recognition
• Explain the motivation

Raw data
• Time-Space-Frequency patterns
• Non-stationary
• Signals measured during specific task

Features extraction
Preprocessing
• Power spectral density (PSD)
• PSD divided into 5 frequency bands for 32 electrodes
• PSD difference between 14 symmetrical electrodes
• Total of 230 input to the network
Normalization
• Baseline subtraction
• Re-scaling into [0.1,0.9] range

Data labeling
• 31 subjects
• DEAP
• 40 one-minute movies (per subject)
• After each movie – self assessment of arousal/valence level
• Total of 31 subject X 40 min movie X 60 PSDs/min = 74400 training sets

Network structure
• Sparce-autoencoder
• Three stacked hidden layers
• Two softmax classifiers
• Fine-tuning (back-prop.)
Unsupervised
Supervised

Autoencoder cost function
2
1 1 1 1
1
ˆ
ˆ
( ) ( || )
2 2
n m n m
i i j ij
i j i j
Cost x x KL W
n

  
   
   
  
2

Kullback-Leibler (KL)
divergence.
Standard function for
measuring difference between
two Bernoulli distributions.
If 𝜌 = ො
𝜌𝑗 KL=0
2
1 1 1 1
1
ˆ
ˆ
( ) ( || )
2 2
n m n m
i i j ij
i j i j
Cost x x KL W
n

  
   
   
  
𝐾𝐿(𝜌||ො
𝜌𝑗) = 𝜌log
𝜌
ො
𝜌𝑗
+ (1 − 𝜌)log
1 − 𝜌
1 − ො
𝜌𝑗
2

• 𝜆 is called the weight decay parameter
• Used to limit the network weights.
2
1 1 1 1
1
ˆ
ˆ
( ) ( || )
2 2
n m n m
i i j ij
i j i j
Cost x x KL W
n

  
   
   
  
2

L1 and L2 Regularization
• Regularization is a process of introducing additional information in
order to prevent overfitting loss function.
• A regression model that uses L1 regularization technique is called
Lasso Regression and model which uses L2 is called Ridge Regression.
The key difference between these two is the penalty term.
• Ridge regression adds “squared magnitude” of coefficient as penalty.
• Gradient descent is simply a method to find the ‘right’ coefficients
through iterative updates using the value of the gradient.
(gradient descent can be used in a simple linear regression)

L1 and L2 Regularization
Regularization is a process of introducing additional information in order to prevent
overfitting loss function.
A regression model that uses L1 regularization technique is called Lasso Regression and
model which uses L2 is called Ridge Regression. The key difference between these two is
the penalty term.
Ridge regression adds “squared magnitude” of coefficient as penalty.
Gradient descent is simply a method to find the ‘right’ coefficients
through iterative updates using the value of the gradient.
(gradient descent can be used in a simple linear regression)
L2 Regularization:
A regression model that uses L1 regularization technique is called Lasso Regression and
model which uses L2 is called Ridge Regression.

• Ridge regression adds “squared magnitude” of coefficient as penalty term to
the loss function.
• If lambda is zero then you can imagine we get back OLS(ordinary Least Square).
However, if lambda is very large then it will add too much weight and it will lead to under-
fitting. Having said that it’s important how lambda is chosen.
• This technique to avoid over-fitting issue.
• Lasso Regression (Least Absolute Shrinkage and Selection Operator) adds
“absolute value of magnitude” of coefficient as penalty term to the loss
function.
• if lambda is zero then we will get back OLS whereas very large value will
make coefficients zero hence it will under-fit.

• The key difference between these techniques is that Lasso shrinks the less
important feature’s coefficient to zero thus, removing some feature altogether.
So, this works well for feature selection in case we have a huge number of
features.
• Traditional methods like cross-validation, stepwise regression to handle
overfitting and perform feature selection work well with a small set of
features but these techniques are a great alternative when we are dealing with
a large set of features.
Batch Normalization
• We normalize the input layer by adjusting and scaling the activations. For
example, when we have features from 0 to 1 and some from 1 to 1000, we
should normalize them to speed up learning.

How does batch normalization work?
To increase the stability of a neural network, batch normalization normalizes the output of a previous
activation layer by subtracting the batch mean and dividing by the batch standard deviation.

• Dropout Neural Network Layer
• Dropout is a technique used to prevent a model from overfitting. Dropout
works by randomly setting the outgoing edges of hidden units (neurons that
make up hidden layers) to 0 at each update of the training phase.

hyper-parameters:
Learning rate, batch size, momentum, and weight decay.
These hyper-parameters act as holds which can be twisted during the training of
the model. For our model to provide best result, we need to find the optimal
value of these hyper-Parameter.
Gradient Descent
Gradient descent is an optimization technique commonly used in training machine learning
algorithms. The main aim of training ML algorithms is to adjust the weights w to minimize
the loss or cost. This cost is measure of how well our model is doing, we represent this cost by
J(w) Thus, by minimizing the cost function we can find the optimal parameters.

CNN (convolutional Neural Network)
• Convolutional layers are the layers where filters are applied to the original image, or to
other feature maps in a deep CNN.
• The name “convolutional neural network” indicates that the network employs a
mathematical operation called convolution .
• Convolution is a specialized kind of linear operation.
• Convolutional networks are simply neural networks that use convolution in place of
general matrix multiplication in at least one of their layers
• Convolutional kernels defined by a width and height (hyper-parameters).
What if we learned the features to detect?
• We need a system that can do Representation Learning (or Feature Learning).
• Representation Learning is a technique that allows a system to automatically find relevant
features for a given task. Replaces manual feature engineering. There are several
techniques for this:
(1)Unsupervised (K-means, PCA, …)
(2)Supervised (Sup. Dictionary learning, Neural Networks!)

Working of CNN:
• Each CNN layer learns filters of increasing complexity.
• The first layers learn basic feature detection filters: edges, corners, etc
• The middle layers learn filters that detect parts of objects. For faces, they might learn to
respond to eyes, noses, etc
• The last layers have higher representations: they learn to recognize full objects, in different
shapes and positions. D example of a CNN working in practice

The Problem with Traditional Neural Networks or MLP
• Traditional neural networks called the multilayer perceptron (MLP).
• If you are not familiar with these, there are hundreds of tutorials on Medium outlining how
MLPs work.
• These are modeled on the human brain, whereby neurons are stimulated by connected
nodes and are only activated when a certain threshold value is reached.
• (problem) : Do not scale well for images
• Ignore the information brought by pixel position and correlation with neighbors
• Cannot handle translations.

Comparison of architecture for MLP and CNN:
• CNN’s are also composed of layers, but those layers are not fully connected.
• They have filters, sets of cube-shaped weights that are applied throughout
the image.
• Each 2D slice of the filters are called kernels. These filters introduce
translation invariance and parameter sharing.

There are several drawbacks of MLP’s
• especially when it comes to image processing. MLPs use one perceptron
for each input (e.g. pixel in an image, multiplied by 3 in RGB case).
• The amount of weights rapidly becomes unmanageable for large images.
For a 224 x 224 pixel image with 3 color channels there are around
150,000 weights that must be trained! As a result, difficulties arise whilst
training and overfitting can occur.
• Clearly, MLPs are not the best idea to use for image processing. One of
the main problems is that spatial information is lost.

Types of Layers in a convolutional neural network:
• There are three types of layers in a convolutional neural network.
(1) convolutional layer(2)pooling layer (3)fully connected layer
(1)convolutional layer:
convolutional layer Convolutional layers are the layers where filters are applied to the
original image, or to other feature maps in a deep CNN. This is where most of the user-
specified parameters are in the network. The most important parameters are the
number of kernels and the size of the kernels.

Pooling layers:
• Pooling layer are similar to convolutional layers, but they perform a specific function
such as max pooling, which takes the maximum value in a certain filter region, or
average pooling, which takes the average value in a filter region.
• These are typically used to reduce the dimensionality of the network.

Connected Layers:
Fully connected layers are placed before the classification output of a CNN and
are used to flatten the results before classification. This is similar to the output
layer of an MLP.

Use of CNN:
The general idea of CNN’s is to intelligently adapt to the properties of images:
• Pixel position and neighborhood
• Elements of interest can appear anywhere in the image
The convolution operation:
We analyze the influence of nearby pixels by using something called a filter.
A filter is exactly what you think it is, in our situation, we take a filter of a size specified by the
user (a rule of thumb is 3x3 or 5x5) and we move this across the image from top left to bottom
right.
For each point on the image, a value is calculated based on the filter using a convolution
operation.

A filter could be related to anything, for pictures of humans, one filter could be associated with
seeing noses, and our nose filter would give us an indication of how strongly a nose seems to
appear in our image, and how many times and in what locations they occur.

Examples of filters or kernels :
After the filters have passed over the image, a feature map is generated for each filter. These
are then taken through an activation function, which decides whether a certain feature is
present at a given location in the image. We can then do a lot of things, such as adding more
filtering layers and creating more feature maps, which become more and more abstract as we
create a deeper CNN.

Example of how convolutions are applied to images using kernel filters:

Padding:
Padding essentially makes the feature maps produced by the filter kernels the
same size as the original image.
This is very useful for deep CNN’s as we don’t want the output to be reduced
so that we only have a 2x2 region left at the end of the network upon which to
predict our result.
Why we Need of Padding?
• If we apply convolutions on a normal image, the result will be down-sampled
by an amount depending on the size of the filter.
• What do we do if we don’t want this to happen? We can use padding.

1x1 convolution :
simply maps an input pixel with all it's channels to an output pixel, not
looking at anything around itself. It is often used to reduce the number
of depth channels, since it is often very slow to multiply volumes with
extremely large depths.
Inception Network: most popular CNNs just stacked convolution layers
deeper and deeper, hoping to get better performance

Recurrent neural network:
A recurrent neural network (RNN) is a class of artificial neural networks where connections
between nodes form a directed graph along a temporal sequence. This allows it to exhibit temporal
dynamic behavior. Derived from feedforward neural networks, RNNs can use their internal state
(memory) to process variable length sequences of inputs. This makes them applicable to tasks such
as unsegmented, connected handwriting recognition or speech recognition.
The term “recurrent neural network” is used indiscriminately to refer to two broad classes of
networks with a similar general structure, where one is finite impulse and the other is infinite
impulse. Both classes of networks exhibit temporal dynamic behavior. A finite impulse recurrent
network is a directed acyclic graph that can be unrolled and replaced with a strictly feedforward
neural network, while an infinite impulse recurrent network is a directed cyclic graph that cannot
be unrolled.
Both finite impulse and infinite impulse recurrent networks can have additional stored states, and
the storage can be under direct control by the neural network. The storage can also be replaced by
another network or graph, if that incorporates time delays or has feedback loops. Such controlled
states are referred to as gated state or gated memory, and are part of long short-term
memory networks (LSTMs) and gated recurrent units. This is also called Feedback Neural
Network.
Recurrent Neural Network remembers the past and it’s decisions are influenced by what it has learnt
from the past. Note: Basic feed forward networks “remember” things too, but they remember things
they learnt during training. For example, an image classifier learns what a “1” looks like during
training and then uses that knowledge to classify things in production.
While RNNs learn similarly while training, in addition, they remember things learnt from prior
input(s) while generating output(s). It’s part of the network.
RNNs can take one or more input vectors and produce one or more output vectors and the output(s)
are influenced not just by weights applied on inputs like a regular NN, but also by a “hidden” state

vector representing the context based on prior input(s)/output(s). So, the same input could produce
a different output depending on previous inputs in the series.
Figure 1: A Recurrent Neural Network, with a hidden state that is meant to carry pertinent
information from one input item in the series to others.
There are following Type of RNN:
Deep RNNs: While it’s good that the introduction of hidden state enabled us to effectively identify the
relationship between the inputs, is there a way we can make a RNN “deep” and gain the multi-level
abstractions and representations we gain through “depth” in a typical neural network?
Figure 2: We can increase depth in three possible places in a typical RNN

Here are four possible ways to add depth. (1) Perhaps the most obvious of all, is to add hidden
states, one on top of another, feeding the output of one to the next. (2) We can also add additional
nonlinear hidden layers between input to hidden state (3) We can increase depth in the hidden to
hidden transition (4) We can increase depth in the hidden to output transition.
Bidirectional RNNs:
Sometimes it’s not just about learning from the past to predict the future, but we also need to look
into the future to fix the past. In speech recognition and handwriting recognition tasks, where there
could be considerable ambiguity given just one part of the input, we often need to know what’s
coming next to better understand the context and detect the present.
Figure 3: Bidirectional RNNs
This does introduce the obvious challenge of how much into the future we need to look
into, because if we have to wait to see all inputs then the entire operation will become
costly. And in cases like speech recognition, waiting till an entire sentence is spoken might
make for a less compelling use case. Whereas for NLP tasks, where the inputs tend to be
available, we can likely consider entire sentences all at once. Also, depending on the
application, if the sensitivity to immediate and closer neighbors is higher than inputs that
come further away, a variant that looks only into a limited future/past can be modeled.
Recursive Neural Networks:
A recurrent neural network parses the inputs in a sequential fashion. A recursive neural network is
similar to the extent that the transitions are repeatedly applied to inputs, but not necessarily in a

sequential fashion. Recursive Neural Networks are a more general form of Recurrent Neural
Networks. It can operate on any hierarchical tree structure. Parsing through input nodes, combining
child nodes into parent nodes and combining them with other child/parent nodes to create a tree like
structure. Recurrent Neural Networks do the same, but the structure there is strictly linear. i.e.
weights are applied on the first input node, then the second, third and so on.
Figure 4: Recursive Neural Net
Long short-term memory (LSTM):
Long Short-Term Memory (LSTM) networks are a modified version of recurrent neural networks,
which makes it easier to remember past data in memory. The vanishing gradient problem of RNN
is resolved here. LSTM is well-suited to classify, process and predict time series given time lags of
unknown duration. It trains the model by using back-propagation. In an LSTM network, three gates
are present:
Figure: 5 LSTM Gates

Input gate — discover which value from input should be used to modify the
memory. Sigmoid function decides which values to let through 0,1. and tanh function gives
weightage to the values which are passed deciding their level of importance ranging from-1 to 1.
Input gates
2. Forget gate — discover what details to be discarded from the block. It is decided by the sigmoid
function. it looks at the previous state(ht-1) and the content input(Xt) and outputs a number
between 0(omit this)and 1(keep this)for each number in the cell state Ct−1.
Forget gate
3. Output gate — the input and the memory of the block is used to decide the
output. Sigmoid function decides which values to let through 0,1. and tanh function gives
weightage to the values which are passed deciding their level of importance ranging from-1 to 1 and
multiplied with output of Sigmoid.
Out put gate

Gated Recurrent Units (GRU):
The GRU is the newer generation of Recurrent Neural networks and is pretty similar to an LSTM.
GRU’s got rid of the cell state and used the hidden state to transfer information. It also only has two
gates, a reset gate and update gate.
GRUcell and it’s gates
Update Gate
The update gate acts similar to the forget and input gate of an LSTM. It decides what information
to throw away and what new information to add.
Reset Gate
The reset gate is another gate is used to decide how much past information to forget.
And that’s a GRU. GRU’s has fewer tensor operations; therefore, they are a little speedier to train
then LSTM’s. There isn’t a clear winner which one is better. Researchers and engineers usually
try both to determine which one works better for their use case.

The GRU is like a long short-term memory (LSTM) with forget gate.
but has fewer parameters
than LSTM, as it lacks an output gate. GRU's performance on certain tasks of polyphonic music
modeling and speech signal modeling was found to be similar to that of LSTM. GRUs have been
shown to exhibit even better performance on certain smaller datasets.
LSTM ’s and GRU’s were created as the solution to short-term memory
The Problem, Short-term Memory
Recurrent Neural Networks suffer from short-term memory. If a sequence is long enough, they’ll
have a hard time carrying information from earlier time steps to later ones. So if you are trying to
process a paragraph of text to do predictions, RNN’s may leave out important information from
the beginning.
During back propagation, recurrent neural networks suffer from the vanishing gradient problem.
Gradients are values used to update a neural networks weight. The vanishing gradient problem is
when the gradient shrinks as it back propagates through time. If a gradient value becomes
extremely small, it doesn’t contribute too much learning.
So in recurrent neural networks, layers that get a small gradient update stops learning. Those are
usually the earlier layers. So because these layers don’t learn, RNN’s can forget what it seen in
longer sequences, thus having a short-term memory.
LSTM’s and GRU’s as a solution
LSTM ’s and GRU’s were created as the solution to short-term memory. They have internal
mechanisms called gates that can regulate the flow of information.

These gates can learn which data in a sequence is important to keep or throw away. By doing that,
it can pass relevant information down the long chain of sequences to make predictions. Almost all
state-of-the-art results based on recurrent neural networks are achieved with these two networks.
LSTM’s and GRU’s can be found in speech recognition, speech synthesis, and text generation.
What is Beam search?
beam search is a heuristic search algorithm that explores a graph by expanding the most promising
node in a limited set. Beam search is an optimization of best-first search that reduces its memory
requirements. Best-first search is a graph search which orders all partial solutions (states)
according to some heuristic. But in beam search, only a predetermined number of best partial
solutions are kept as candidates.
The beam search algorithm selects multiple alternatives for an input sequence at each timestep
based on conditional probability. The number of multiple alternatives depends on a parameter
called Beam Width B. At each time step, the beam search selects B number of best alternatives
with the highest probability as the most likely possible choices.
Uses:

A beam search is most often used to maintain tractability in large systems with insufficient amount
of memory to store the entire search tree. For example, it has been used in many machine
translation systems. (the state of the art now primarily uses neural machine translation based
methods). To select the best translation, each part is processed, and many different ways of
translating the words appear. The top best translations according to their sentence structures are
kept, and the rest are discarded. The translator then evaluates the translations according to a given
criterion, choosing the translation which best keeps the goals.
Reinforcement learning and Frame work: is a decision-making system its ability to takes
decision with the help of reward and penalty over environment. Agent is a program code and also
called decision making. RL is based on self-learning phenomenon that have not prior knowledge
about system.
After offset of trial and error runs, it should learn the best policy, which is the sequence of actions
that maximize the total reward. This is achieved entire goals through optimal action.
It offers a general structure and several methods to make it improve its behavior. RL is not That
In case of large search spaces or large data sets our existing learning algorithm required large
storage or look up table. Matter is that how to memorize or reduce or optimize look-up-table.
We are analyzing and study that how to optimize look up table. It can be possible with the help of
DCT. It is also helpful for exponential data storage in matrix form.
In RL, the optimal action output is not provided to the learner, which is in contrast to supervised
learning methods. Therefore, the learner is able to change current policy and commutative rewards
during trial and error interaction. In recent years, among most reactive control methods,
ENVIRONMENT
AGENT
Action
State
Fig.1 The agent interacts with environment at any
state of the environment, the agent takes action
that changes the state and returns a reward.

reinforcement learning has been broadly applied into robot navigation field in unknown
environment because of its self-learning and on line learning abilities. RL is unsupervised or on-
line learning method, in which agent is given feedback via reward and penalty over environment.
The basic idea is that action correlated with high reward and due to higher iterative probability
required much storage space, it is required for mapping the situation, applied by hidden Marko
model (HMM). While those correlated with low reward have lower iterative probability, Used
method as a incremental and real time learning method.
Reinforcement learning focuses on, how to agent learn without prior knowledge of the system. We
can say that it may applied over un-certain environment through commutative reward, it should be
maximized. In consecutive learning agent should require customized expected reward which help
for selection of the optimal policy and chooses optimal policy among them. After select policy
apply action over environment.
There are many unsolved problems that computers could solve if the appropriate software existed.
Reinforcement learning (RL) is an area of machine learning concerned with how software
agents ought to take actions in an environment in order to maximize the notion of cumulative
reward. Reinforcement learning is one of three basic machine learning paradigms,
alongside supervised learning and unsupervised learning.
Reinforcement learning differs from supervised learning in not needing labelled input/output pairs
be presented, and in not needing sub-optimal actions to be explicitly corrected. Instead the focus
is on finding a balance between exploration (of uncharted territory) and exploitation (of current
knowledge).
Reinforcement learning, due to its generality, is studied in many other disciplines, such as game
theory, control theory, operations research, information theory, simulation-based
optimization, multi-agent systems, swarm intelligence, statistics and genetic algorithms.
reinforcement learning is called approximate dynamic programming, or neuro-dynamic
programming. The problems of interest in reinforcement learning have also been studied in
the theory of optimal control, which is concerned mostly with the existence and characterization
of optimal solutions, and algorithms for their exact computation, and less with learning or
approximation, particularly in the absence of a mathematical model of the environment.

In economics and game theory, reinforcement learning may be used to explain how equilibrium
may arise under bounded rationality.
Markov decision process(MDP):
A Markov decision process (MDP) is a discrete time stochastic control process. It provides a
mathematical framework for modeling decision making in situations where outcomes are
partly random and partly under the control of a decision maker. MDPs are useful for
studying optimization problems solved via dynamic programming and reinforcement learning.
MDPs were known also Dynamic Programming and Markov Processes. They are used in many
disciplines, including robotics, automatic control, economics and manufacturing. The name of
MDPs comes from the Russian mathematician Andrey Markov as they are an extension of
the Markov chains.
A stochastic process has the Markov property if the conditional probability distribution of future
states of the process (conditional on both past and present values) depends only upon the present
state; that is, given the present, the future does not depend on the past. A process with this property
is said to be Markovian or a Markov process. The most famous Markov process is a Markov
chain. Brownian motion is another well-known Markov process.
A Bellman equation, named after Richard E. Bellman, is a necessary condition for optimality
associated with the mathematical optimization method known as dynamic programming.
However, the term 'Bellman equation' usually refers to the dynamic
programming equation associated with discrete-time optimization problems.
To understand the Bellman equation, several underlying concepts must be understood. First, any
optimization problem has some objective: minimizing travel time, minimizing cost, maximizing profits,
maximizing utility, etc. The mathematical function that describes this objective is called the objective
function.
Dynamic programming breaks a multi-period planning problem into simpler steps at different points in
time. Therefore, it requires keeping track of how the decision situation is evolving over time. The
information about the current situation that is needed to make a correct decision is called the state.
For example, to decide how much to consume and spend at each point in time, people would need to
The variables chosen at any given point in time are often called the control variables. For example,
given their current wealth, people might decide how much to consume now. Choosing the control
variables now may be equivalent to choosing the next state; more generally, the next state is affected
by other factors in addition to the current control. For example, in the simplest case, today's wealth

(the state) and consumption (the control) might exactly determine tomorrow's wealth (the new state),
though typically other factors will affect tomorrow's wealth too.
The dynamic programming approach describes the optimal plan by finding a rule that tells what the
controls should be, given any possible value of the state.
A Bellman equation is represented as :
Value Iteration and Policy Iteration:
In previous sections the environment and the reinforcement function are discussed. However, the issue of
how the agent learns to choose “good” actions or even how we might measure the utility of an action is not
explained. First, two terms are defined. Policy determines which action should be performed in each state;
a policy is a mapping from states to actions. The value of a state is defined as the sum of the reinforcements
received when starting in that state and following some fixed policy to a terminal state. The optimal policy
would therefore be the mapping from states to actions that maximizes the sum of the reinforcements.
The value function is a mapping from states to state values and can be approximated using any type of
function approximator (e.g., multilayered perceptron, memory-based system, radial basis functions, look-
up table, etc.).
An example of a value function can be seen using a simple Markov decision process with 16 states. The
state space can be visualized using a 4x4 grid in which each square represents a state.
The reinforcement function is ( -1) everywhere (i.e., the agent receives a reinforcement of (-1) on each
transition). There are 4 actions possible in each state: north, south, east, and west. The goal states are the
upper left corner and the lower right corner. The value function for the random policy is shown in Figure
1. For each state the random policy randomly chooses one of the four possible actions. The numbers in the
states represent the expected values of the states. For example, when starting in the lower left corner and
following a random policy, on average there will be 22 transitions to other states before the terminal state
is reached.

The optimal value function is shown in Figure 2. Again, starting in the lower left corner, calculating the
sum of the reinforcements when performing the optimal policy (the policy that will maximize the sum of
the reinforcements), the value of that state is -3 because it takes only 3 transitions to reach a terminal state.
If we are given the optimal value function, then it becomes a trivial task to extract the optimal policy. For
example, one can start in any state in Figure 2 and simply choose the action that maximizes the immediate
reinforcement received. In other words, one can perform a one level deep breadth-first search over actions
to find the action that will maximize the immediate reward. The optimal policy for the value function shown
in Figure 2 is given in Figure 3.
Approximating the Value Function:
Reinforcement learning is a difficult problem because the learning system may perform an action and not
be told whether that action was good or bad. For example, a learning auto-pilot program might be given
control of a simulator and told not to crash. It will have to make many decisions each second and then, after
acting on thousands of decisions, the aircraft may crash due to …….. What should the system learn from
this experience? Which of its many actions was responsible for the crash? Assigning blame to individual
0 -14 -20 -22
-14 -18 -22 -20
-20 -22 -18 -14
-22 -20 -14 0
0 -1 -2 -3
-1 -2 -3 -2
-2 -3 -2 -1
-3 -2 -1 0
Figure 1
Figure 2
Figure 3

actions is the problem that makes reinforcement learning difficult. Surprisingly, there is a solution to this
problem. It is based on a field of mathematics called dynamic programming, and it involves just two basic
principles. First, if an action causes something bad to happen immediately, such as crashing the plane, then
the system learns not to do that action in that situation again. So whatever action the system performed one
millisecond before the crash, it will avoid doing in the future. But that principle doesn’t help for all the
earlier actions which didn’t lead to immediate disaster.
The second principle is that if all the actions in a certain situation lead to bad results, then that situation
should be avoided. So if the system has experienced a certain combination of altitude and airspeed at many
different times, whereby trying a different action each time, and all actions led to something bad, then it
will learn that the situation itself is bad. This is a powerful principle, because the learning system can now
learn without crashing. In the future, any time it chooses an action that leads to this particular situation, it
will immediately learn that particular action is bad, without having to wait for the crash. By using these two
principles, a learning system can learn to fly a plane, control a robot, or do any number of tasks. It can first
learn on a simulator, then fine tune on the actual system. This technique is generally referred to as dynamic
programming, and a slightly closer analysis will reveal how dynamic programming can generate the optimal
value function.
Difference between policy iteration and Value iteration:
In policy iteration algorithms, you start with a random policy, then find the value function of that
policy (policy evaluation step), then find a new (improved) policy based on the previous value
function, and so on. In this process, each policy is guaranteed to be a strict improvement over the
previous one (unless it is already optimal). Given a policy, its value function can be obtained using
the Bellman operator.
In value iteration, you start with a random value function and then find a new (improved) value
function in an iterative process, until reaching the optimal value function. Notice that you can
derive easily the optimal policy from the optimal value function. This process is based on the
optimality Bellman operator.
In some sense, both algorithms share the same working principle, and they can be seen as two
cases of the generalized policy iteration. However, the optimality Bellman operator contains a max
operator, which is nonlinear and, therefore, it has different features. In addition, it's possible to use
hybrid methods between pure value iteration and pure policy iteration.
Actor-Critics aim to take advantage of all the good stuff from both value-based and
policy-based while eliminating all their drawbacks. And how do they do this? The principal

idea is to split the model in two: one for computing an action based on a state and another
one to produce the Q values of the action.
Figure: Actor-critic model
Q-learning, SARSA
Q-Learning: Q-learning is a form of model free reinforcement learning. It provides agent with the
capability of learning to act optimally in Markov domains by experiencing action consequences. Agent
selects optimal policy through iteration (Q function), Q learning usually stores Q value (S, A) relative with
every state action in a lookup table . In Q learning there are learning steps repeat in every episode till nth
time:
Evaluation of Q value:
Q values of each agent are evaluated by the sum of the rewards which the agent obtains during the previous
one episode using this equation.
Q(sn ,an)=rn+1+γ max an+1Q(sn+1,an+1)
And simply use this as an assignment to update Q(sn ,an) When in state sn, we choose action an by one of
the stochastic strategies we saw earlier , which returns a reward rn+1 and takes us to state sn+1. We then
update the value of previous action as.
Q`(sn ,an)=rn+1+γ max an+1Q`(sn+1,an+1)
Where the des denote the value is an estimate Q`(sn+1,an+1) is a later value and has higher chance of being
correct
γ : discounted rate parameter that reduces the influence of future expected rewards(0<γ<1)
r : reward
sn: next step
an: action taken in
s0: initial step
α= learning rate
a0=initial action
Q value is updated whenever an agent takes an action in the individual learning, Q values at and shortly
after beginning rather of the episode is rather different from those at the end of the episode. Even if a new
episode begins with the Q values at the end of the previous episode are not necessarily taken and the same
rewards are not necessarily obtained. The rewards obtained by using the Q values at and shortly after the
beginning are considered to have only a little relation with the Q values at the end. Thus such rewards are

n
Qi(s, a) Qbest
(s ,a)
discounted and the discounted and the discounted results are summed up. Therefore, we define the evaluated
value E for the Q value at the end each episode .
Where N is the number of actions in the episode, rk is the reward for the kth
action, and d(<1) is the discount
parameter , The definition (1) could bring that the larger the evaluated value E for Q-values is the superior
the Q-value are:
Update Q value In order to update Q-values through exchanging the information among the agents, Best
state action value method, If the superior Q values are copied to those for each of the other agents, each
agent can improve its Q-values in future episodes, according to this idea the Q value Qi(s ,a) of agent I for
action a in state s is updated by
Where Qbest
(s ,a) is the Q-value which is evaluated superior to that of any other agent , and is called the
best Q-value.
Average state action value method in best, the Q-values which are not best are discarded, and replaced
with the best Q-values. However, they are not necessarily worthless, because they are updated by agents
through the individual learning. Optimal Q-values may not be found [5]. These drawbacks of BEST, in AVE
each agent updates its Q-values by averaging its current Q-value and the best Q-value for each state
action.
SARSA (State Action Reward State Action):
SARSA algorithm is a slight variation of the popular Q-Learning algorithm. For a learning agent
in any Reinforcement Learning algorithm it’s policy can be of two types:-
1. On Policy: In this, the learning agent learns the value function according to the current
action derived from the policy currently being used.
2. Off Policy: In this, the learning agent learns the value function according to the action
derived from another policy.
Q-Learning technique is an Off-Policy technique and uses the greedy approach to learn the Q-
value. SARSA technique, on the other hand, is an On Policy and uses the action performed by
the current policy to learn the Q-value. the difference of the update statements for each
technique: -
Here, the update equation for SARSA depends on the current state, current action,
reward obtained, next state and next action. This observation lead to the naming of the
learning technique as SARSA stands for State Action Reward State Action which
symbolizes the tuple (s, a, r, s’, a’).

Support Vector Machine (SVM):
The objective of the support vector machine algorithm is to fined a hyperplane in
an N-dimensional space that classify the data point.
support-vector machines (SVMs) are supervised learning models with associated
learning algorithms that analyze data used for classification and regression analysis.
Given a set of training examples, each marked as belonging to one or the other of
two categories, an SVM training algorithm builds a model that assigns new examples
to one category or the other, making it a non-probabilistic binary linear
classifier (although methods such as Platt scaling exist to use SVM in a probabilistic
classification setting). An SVM model is a representation of the examples as points
in space, mapped so that the examples of the separate categories are divided by a
clear gap that is as wide as possible. New examples are then mapped into that same
space and predicted to belong to a category based on the side of the gap on which
they fall.
In addition to performing linear classification, SVMs can efficiently perform a non-
linear classification using what is called the kernel trick, implicitly mapping their
inputs into high-dimensional feature spaces.
When data are unlabeled, supervised learning is not possible, and an unsupervised
learning approach is required, which attempts to find natural clustering of the data to
groups, and then map new data to these formed groups.

Possible hyperplanes
To separate the two classes of data points, there are many possible hyperplanes that
could be chosen. Our objective is to find a plane that has the maximum margin, i.e
the maximum distance between data points of both classes. Maximizing the margin
distance provides some reinforcement so that future data points can be classified
with more confidence.
Hyperplanes and Support Vectors:
Hyperplanes in 2D and 3D feature space
Hyperplanes are decision boundaries that help classify the data points. Data points
falling on either side of the hyperplane can be attributed to different classes. Also,
the dimension of the hyperplane depends upon the number of features. If the
number of input features is 2, then the hyperplane is just a line. If the number of

input features is 3, then the hyperplane becomes a two-dimensional plane. It
becomes difficult to imagine when the number of features exceeds 3.
Support vectors are data points that are closer to the hyperplane and influence the
position and orientation of the hyperplane. Using these support vectors, we
maximize the margin of the classifier. Deleting the support vectors will change the
position of the hyperplane. These are the points that help us build our SVM.
Large Margin Intuition
In logistic regression, we take the output of the linear function and squash the value
within the range of [0,1] using the sigmoid function. If the squashed value is greater
than a threshold value (0.5) we assign it a label 1, else we assign it a label 0. In
SVM, we take the output of the linear function and if that output is greater than 1,
we identify it with one class and if the output is -1, we identify is with another class.
Since the threshold values are changed to 1 and -1 in SVM, we obtain this
reinforcement range of values ([-1,1]) which acts as margin.
Cost Function and Gradient Updates
In the SVM algorithm, we are looking to maximize the margin between the data
points and the hyperplane. The loss function that helps maximize the margin is
hinge loss.

Hinge loss function n (function on left can be represented as a function on the
right)
The cost is 0 if the predicted value and the actual value are of the same sign. If they
are not, we then calculate the loss value. We also add a regularization parameter the
cost function. The objective of the regularization parameter is to balance the margin
maximization and loss. After adding the regularization parameter, the cost functions
look as below.
Loss function for SVM
Now that we have the loss function, we take partial derivatives with respect to the
weights to find the gradients. Using the gradients, we can update our weights.
Gradients
When there is no misclassification, i.e our model correctly predicts the class of our
data point, we only have to update the gradient from the regularization parameter.
Gradient Update — No misclassification
When there is a misclassification, i.e our model make a mistake on the prediction
of the class of our data point, we include the loss along with the regularization
parameter to perform gradient update.

Gradient Update Misclassification
Bayesian learning:
So far, we have discussed Bayes’ theorem and gained an understanding of how
we can apply Bayes’ theorem to test our hypotheses. However, when using
single point estimation techniques such as MAP, we will not be able to exploit
the full potential of Bayes’ theorem. We used single values to explain each term
in Bayes’ theorem to simplify my explanation of Bayes’ theorem.
Bayesian learning, we are dealing with random variables that have probability
distributions. Let us try to understand why using exact point estimations can be
misleading in probabilistic concepts.
Bayes’ Theorem:
Bayes’ theorem describes how the conditional probability of an event or a
hypothesis can be computed using evidence and prior knowledge. It is similar to
concluding that our code has no bugs given the evidence that it has passed all the
test cases, including our prior belief that we have rarely observed any bugs in our
code. However, this intuition goes beyond that simple hypothesis test where there
are multiple events or hypotheses involved (let us not worry about this for the
moment).

Maximum a Posteriori (MAP):
We can use MAP to determine the valid hypothesis from a set of hypotheses.
According to MAP, the hypothesis that has the maximum posterior probability
is considered as the valid hypothesis.
Binomial Likelihood:
The likelihood for the coin flip experiment is given by the probability of
observing heads out of all the coin flips given the fairness of the coin.
As we have defined the fairness of the coins using the probability of observing
heads for each coin flip, we can define the probability of observing heads or
tails given the fairness of the coin P(y|theta) where y = 1 for observing heads
and y = 0, observing tails. Accordingly:

Now that we have defined two conditional probabilities for each outcome
above, let us now try to find the P(Y=y| theta) joint probability of observing
heads or tails:
We can rewrite the above expression in a single expression as follows:
The above equation represents the likelihood of a single test coin flip
experiment. Interestingly, the likelihood function of the single coin flip
experiment is similar to the Bernoulli probability distribution. The Bernoulli
distribution is the probability distribution of a single trial experiment with only
two opposite outcomes.
Binomial probability distribution as shown below:
Beta Prior Distribution
The prior distribution is used to represent our belief about the hypothesis based on
our past experiences. We can choose any distribution for the prior, if it represents
our belief regarding the fairness of the coin. For this example, we use Beta
distribution to represent the prior probability distribution as follows:

Application of machine learning in computer vision:
Machine learning has improved computer vision about recognition and tracking. It
offers effective methods for acquisition, image processing, and object focus which
are used in computer vision. There are following field where computer Vision
applied:
1. Image Classification
2. Image Classification with Localization
3. Object Detection
4. Object Segmentation
5. Image Style Transfer
6. Image Colorization
7. Image Reconstruction
8. Image Super-Resolution
9. Image Synthesis
10.Other Problems
Image Classification
Image classification involves assigning a label to an entire image or photograph.
This problem is also referred to as object classification and perhaps more generally
as image recognition, although this latter task may apply to a much broader set of
tasks related to classifying the content of images.
Some examples of image classification include:
• Labeling an x-ray as cancer or not (binary classification).
• Classifying a handwritten digit (multiclass classification).
• Assigning a name to a photograph of a face (multiclass classification).
Image Classification with Localization
Image classification with localization involves assigning a class label to an image
and showing the location of the object in the image by a bounding box (drawing a
box around the object).
This is a more challenging version of image classification.

Some examples of image classification with localization include:
• Labeling an x-ray as cancer or not and drawing a box around the cancerous region.
• Classifying photographs of animals and drawing a box around the animal in each
scene.
Object Detection
Object detection is the task of image classification with localization, although an
image may contain multiple objects that require localization and classification.
This is a more challenging task than simple image classification or image
classification with localization, as often there are multiple objects in the image of
different types.
Often, techniques developed for image classification with localization are used and
demonstrated for object detection.
Some examples of object detection include:
• Drawing a bounding box and labeling each object in a street scene.
• Drawing a bounding box and labeling each object in an indoor photograph.
• Drawing a bounding box and labeling each object in a landscape.
Object Segmentation
Object segmentation, or semantic segmentation, is the task of object detection
where a line is drawn around each object detected in the image. Image
segmentation is a more general problem of spitting an image into segments. Object
detection is also sometimes referred to as object segmentation. Unlike object
detection that involves using a bounding box to identify objects, object
segmentation identifies the specific pixels in the image that belong to the object. It
is like a fine-grained localization.
Application of machine learning in computer
speech processing:
Machine learning uses iterative algorithms to learn from data and allows
the computer to find information, hidden values that are not explicitly

programmed. Machine Learning systems can quickly apply knowledge and
training from large datasets to perform face recognition, speech recognition.
Introducing speech recognition digital assistants
Digital assistants are designed to help people perform or complete basic tasks and
respond to queries. With the ability to access information from vast databases and
various digital sources, these robots help to solve problems in real time, enhancing
the user experience and human productivity.
Popular digital assistants, include:
• Amazon’s Alexa
• Apple’s Siri
• Google’s Google Assistant
• Microsoft’s Cortana
Applications of speech recognition technology:
1. In the workplace
Speech recognition technology in the workplace has evolved into incorporating
simple tasks to increase efficiency, as well as beyond tasks that have traditionally
needed humans, to be performed.
Examples of office tasks digital assistants are, or will be, able to perform:
• Search for reports or documents on your computer
• Create a graph or tables using data
• Dictate the information you want to be incorporated into a document
• Print documents on request
• Start video conferences
• Schedule meetings
• Record minutes
• Make travel arrangements

2. In banking
The aim of the banking and financial industry is for speech recognition to reduce
friction for the customer.8
Voice-activated banking could largely reduce the need for
human customer service, and lower employee costs. A personalized banking
assistant could in return boost customer satisfaction and loyalty. How speech
recognition could improve banking:
• Request information regarding your balance, transactions, and spending habits
without having to open your cell phone
• Make payments
• Receive information about your transaction history
4. In Healthcare
In an environment where seconds are crucial and sterile operating conditions are a
priority, hands-free, immediate access to information can have a significantly
positive impact on patient safety and medical efficiency.
Benefits include:
• Quickly finding information from medical records
• Nurses can be reminded of processes or given specific instructions
• Nurses can ask for administrative information, such as the number of patients on a
floor and the number of available units
• At home, parents can ask for common symptoms of diseases, when they should go
to the doctor, and how to look after a sick child
• Less paperwork
• Less time inputting data
• Improved workflows
The most significant concern using speech recognition in healthcare is the content
the digital assistant has access to. It has been recognized that the content will need
to be supplied and validated by recognized medical institutions, in order for it to be
a viable option in this field.
Application of machine learning in computer natural language processing:
1. Text Classification and Categorization

Text classification is an essential part in many applications, such as web
searching, information filtering, language identification, readability
assessment, and sentiment analysis. Neural networks are actively used for
these tasks.
2. Semantic Parsing and Question Answering
Question Answering systems automatically answer different types of questions
asked in natural languages including definition questions, biographical questions,
multilingual questions, and so on. Neural networks usage makes it possible to
develop high performing question answering systems.
3. Language Generation and Multi-document Summarization
Natural language generation has many applications such as automated writing of
reports, generating texts based on analysis of retail sales data, summarizing
electronic medical records, producing textual weather forecasts from weather data,
and even producing jokes.
4. Machine Translation
Machine translation software is used around the world despite its limitations. In
some domains, the quality of translation is not good. To improve the results
researchers, try different techniques and models, including the neural network
approach. The purpose of Neural-based Machine Translation for Medical Text
Domain study is to inspect the effects of different training methods on a Polish-
English machine translation system used for medical data. To train neural and
statistical network-based translation systems.
5. Spell Checking
Most text editors let users check if their text contains spelling mistakes. Neural
networks are now incorporated into many spell-checking tools.
6. Character Recognition
Character Recognition systems also have numerous applications like receipt
character recognition, invoice character recognition, check character recognition,
legal billing document character recognition, and so on.

Linear Regression

More Related Content

What's hot (20)

Similar to Linear Regression (20)

Recently uploaded (20)

Linear Regression