9/25/2024

MODULE 1
INTRODUCTION ON DEEP
LEARNING

DR. SHIVA PRASAD KM

ASSOCIATE PROFESSOR
DEPARTMENT OF CSE
RYMEC BALLARI
M: 7899964163

OVERVIEW
• INTRODUCTION OF DEEP LEARNING
• DIFFERENTIATE BETWEEN AI VS ML VS DEEP LEARNING
• HISTORY OF DEEP LEARNING
• DEEP LEARNING MODEL
• HISTORICAL TRENDS IN DEEP LEARNING
• MACHINE LEARNING BASICS
• SUPERVISED LEARNING
• UNSUPERVISED LEARNING
• REINFORCEMENT LEARNING
• THE TASK T
• THE PERFORMANCE P
• THE EXPERIENCE E
Dr. Shiva Prasad KM Associate Professor CSE Department RYMEC Ballari M: 7899964163
 9/25/2024

INTRODUCTION OF DEEP LEARNING

■ Deep learning is a subfield of machine learning that uses artificial neural networks to learn
from data.

■ It's a powerful tool for solving complex problems in various domains, including image
recognition, natural language processing, and robotics.

■ Deep learning requires large amount of data to train and test the model which inturn
increases the time complexity and also requires high computational power.

■ Techniques like Convoluntional neural networks (CNNs), Recurrent Neural Networks

(RNNs), and Transformers are common in deep learning.

INTRODUCTION OF DEEP LEARNING (EXAMPLE)

 9/25/2024

AI VS ML VS DEEP LEARNING
■ Artificial Intelligence (AI): ■ Machine Learning (ML): ■ Deep Learning (DL):
– Definition: AI is the broad– Definition: ML is a subset of– Definition: DL is a subset of
field of creating intelligent AI that focuses on building ML that uses deep neural
machines that can mimic or algorithms that enable networks with multiple layers
simulate human intelligence. machines to learn from data to model complex data
– Example: Virtual assistants and improve their performance patterns.
like Siri and Alexa that can over time without explicit– Example: Self-driving cars
understand voice commands programming. that use deep learning models
and provide relevant responses– Example: A spam email filter to recognize road signs,
are AI systems. that learns from user pedestrians, and other vehicles
behaviour to identify spam in real time.
emails.

INTRODUCTION OF DEEP LEARNING

 9/25/2024

HISTORY OF DEEP LEARNING

■ Early Foundations (1940s - 1970s):

– 1943: The concept of artificial neurons was introduced by Warren McCulloch and Walter Pitts, who proposed a
mathematical model of how biological neurons might function, laying the groundwork for neural networks.

– 1958: Frank Rosenblatt developed the perceptron, an early model for binary classifiers that could learn simple
patterns. However, perceptron were limited and could not solve non-linear problems.

■ Neural Network Winter (1970s - 1980s):

– Interest in neural networks declined during this period, partly due to limitations such as the inability to handle
non-linear problems. A famous book by Marvin Minsky and Seymour Papert in 1969 highlighted these
limitations.

– Back propagation (1986): A breakthrough came when Geoffrey Hinton, David Rumelhart, and Ronald
Williams rediscovered and popularized the back propagation algorithm, which allowed for training multi-
layer neural networks, helping to overcome previous challenges.

HISTORY OF DEEP LEARNING

■ Revival of Neural Networks (1990s):

– With increased computing power and better algorithms, neural networks started to regain popularity.
Yann LeCun’s work on Convoluntional Neural Networks (CNNs), particularly for digit recognition,
showed the promise of deep networks in real-world tasks.

■ Deep Learning Era (2000s - Present):

– 2006: Geoffrey Hinton introduced the concept of deep belief networks, which led to a resurgence of
interest in deep architectures. This marked the beginning of modern deep learning, where deeper
neural networks with more layers became feasible due to improved algorithms and hardware.

– 2012: Deep learning gained significant attention when Alex Krizhevsky, Ilya Sutskever, and
Geoffrey Hinton used CNNs to win the ImageNet competition by a large margin, revolutionizing
computer vision.
 9/25/2024

HISTORY OF DEEP LEARNING

■ Ian Good fellow and GANs (2014):

– In 2014, Ian Good fellow introduced Generative Adversarial Networks (GANs), a groundbreaking
deep learning technique where two neural networks (a generator and a discriminator) compete
against each other to produce highly realistic data, such as images. GANs opened new possibilities in
areas like image synthesis, video generation, and even art.

■ Deep Learning Today:

– Deep learning is now at the core of advancements in various domains, including natural language
processing (e.g., Open AI's GPT series), autonomous driving, healthcare, and robotics. Transformers,
attention mechanisms, and transfer learning are key developments that continue to push the
boundaries of what deep learning can achieve.

DEEP LEARNING MODEL

■ A deep learning model refers to a type of machine learning architecture composed of multiple layers of
neural networks, designed to automatically learn patterns from data.

■ The various key factors of deep learning model are:

■ Layers:
– Input Layer: Receives the raw data (e.g., images, text, etc.).

– Hidden Layers: These are the intermediate layers that learn to extract features from the data. The
more hidden layers, the "deeper" the model. Hidden layers perform various transformations of the
input data using weights and activation functions.

– Output Layer: Produces the final result, like a classification label or regression value.
 9/25/2024

DEEP LEARNING MODEL

■ Weights:
– Each connection between neurons has an associated weight. These weights are learned during
training and are updated using optimization techniques like gradient descent.
■ Activation Functions:
– Introduce non-linearity into the model. Common activation functions include ReLU (Rectified Linear
Unit), sigmoid, and tanh.
■ Loss Function:
– A measure of how far the predicted output is from the actual target output. During training, the model
tries to minimize the loss function (e.g., mean squared error for regression or cross-entropy for
classification).
■ Back propagation and Gradient Descent:
– Back propagation: A technique to update weights by calculating the gradient of the loss function
concerning the weights.
– Gradient Descent: An optimization algorithm that adjusts the weights based on the gradients to
minimize the loss function.

DEEP LEARNING MODEL

 9/25/2024

MACHINE LEARNING BASICS

■ Machine learning (ML) algorithms are mathematical models that allow computers to learn
patterns from data and make predictions or decisions without explicit programming.

■ There are three different types of Machine learning algorithms they are

– Supervised Machine learning

– Unsupervised Machine learning

– Reinforcement Machine learning

SUPERVISED MACHINE LEARNING

■ Definition: In supervised learning, the algorithm is trained on a labeled dataset, meaning that each input data point has an
associated correct output.

■ Goal: The model learns to map input data to the correct output (label) and can generalize to predict outcomes on new, unseen
data.

■ Common Algorithms:

– Linear Regression: Used for predicting continuous values (e.g., predicting house prices).

– Logistic Regression: Used for binary classification problems (e.g., spam detection).

– Decision Trees: A tree-like model that makes decisions based on feature values.

– Support Vector Machines (SVM): Finds the optimal boundary (hyper plane) to separate data points into different classes.

– K-Nearest Neighbors (KNN): Classifies data points based on the majority label of the nearest neighbors.

– Example: Predicting whether an email is spam or not, based on labeled examples.

 9/25/2024

SUPERVISED MACHINE LEARNING

UNSUPERVISED MACHINE LEARNING

■ Definition: In unsupervised learning, the algorithm is given data without labeled outcomes. The goal
is to find hidden patterns or structures within the data.

■ Goal: The algorithm groups data or finds associations, without predefined labels.

■ Common Algorithms:

– K-Means Clustering: Groups data into clusters based on similarities.

– Hierarchical Clustering: Builds a tree-like structure of clusters.

– Principal Component Analysis (PCA): Reduces the dimensionality of the data while preserving most
of the variance.

– Example: Grouping customers based on purchasing behaviour without prior knowledge of customer
categories.
 9/25/2024

UNSUPERVISED MACHINE LEARNING

REINFORCEMENT MACHINE LEARNING

■ Definition: Reinforcement learning involves an agent that learns by interacting with its environment
and receiving feedback through rewards or penalties.

■ Goal: The agent learns to make a sequence of decisions to maximize cumulative rewards.

■ Common Algorithms:

– Q-Learning: A value-based approach where the agent learns the value of taking certain actions in
given states.

– Deep Q-Networks (DQN): Combines Q-learning with deep neural networks to handle complex
environments.

– Example: Teaching a robot to navigate a maze by rewarding it for reaching the goal and penalizing
it for hitting obstacles.
 9/25/2024

REINFORCEMENT MACHINE LEARNING

THE TASK T
 In the context of machine learning, Task T refers to the specific problem that a machine learning
model is designed to solve.

 A task T can be defined as any problem where the goal is to map inputs to outputs in a way that
improves performance over time with experience.
■ Types of Tasks in Machine Learning
– Classification Tasks
– Regression Tasks
– Clustering Tasks
– Dimensionality Reduction Tasks
– Generative Tasks
– Sequence-to-Sequence Tasks
 9/25/2024

THE PERFORMANCE P
■ Performance P in machine learning, is a measure of how well a model accomplishes its task T.

■ It is evaluated using various metrics, depending on the type of problem, and is influenced by factors
such as over fitting, under fitting, and generalization.

■ Achieving optimal performance requires balancing model complexity and ensuring that it generalizes
well to new data.
■ Key Aspects of Performance P
– Performance Metrics
– Training and Testing
– Over fitting and Under fitting
– Generalization
– Optimization and Learning

THE EXPERIENCE E
■ In machine learning, Experience (E) refers to the data and feedback a model uses to learn and
improve its performance on a task T.
■ Key Aspects of Experience E:
■ Training Data:
■ Learning from Data:
■ Feedback and Iteration
■ Generalization from Experience
■ Data Representation
■ Transfer Learning
 9/25/2024

LINEAR REGRESSION MODEL

■ Linear regression is one of the most fundamental machine learning algorithms used for predicting
continuous values based on input features.

■ Linear Regression is a basic but essential algorithm for predicting continuous values. By learning
the best-fit line through the data, it captures the linear relationships between input features and the
target variable, making it useful for tasks like price prediction, demand forecasting, and more.

■ Example:
■ Problem: Predicting pizza price based on the size of the pizza.
– Input (X): The size of the pizza (e.g., diameter).
– Output (Y): The price of the pizza (continuous value).

LINEAR REGRESSION MODEL

■ Model:
■ The linear regression model assumes a linear relationship between the input X and the output Y. The
model can be written as:

■ Where:
– w is the weight (slope of the line), (How much Y Changes for the unit change in X)
– b is the bias (y-intercept of the line), (The value of Y when X=0)
– X is the input (size of the pizza)- Independent Variable ,
– Y is the predicted output (price of the Pizza )-Dependent Variable.
■ Training the Model
■ Objective: The goal of linear regression is to find the best values for w and b that minimize the
difference between the predicted prices and the actual prices in the training data.
 9/25/2024

LINEAR REGRESSION MODEL

■ Loss Function: The model minimizes the mean squared error (MSE), which is the average of the
squared differences between the predicted values y and the actual values Y.

■ Gradient Descent: Linear regression uses optimization techniques like gradient descent to adjust
the weight w and bias b iteratively to minimize the loss function.

SUPERVISED MACHINE LEARNING: PROBABILISTIC SUPERVISED LEARNING

■ Probabilistic supervised learning in machine learning involves building models that predict the
probability distribution of target variables given input features.

■ This approach incorporates uncertainty into predictions, allowing the model to estimate not just a
single value but a range of possible outcomes with associated probabilities.
 9/25/2024

SUPERVISED MACHINE LEARNING: PROBABILISTIC SUPERVISED LEARNING

■ Key Concepts of Probabilistic Supervised Learning
1. Probabilistic Models: These models predict the conditional probability distribution P(y∣x) of the
target variable y given input features x. This allows the model to quantify uncertainty in its
predictions.
2. Likelihood Function: In probabilistic learning, the model parameters are typically estimated by
maximizing the likelihood function, which measures how well the model predicts the observed data.
3. Bayesian Interpretation: Probabilistic models often have a Bayesian interpretation, where prior
beliefs about the model parameters are updated with observed data to form posterior distributions.
4. Loss Functions: Probabilistic models use loss functions derived from probability distributions, such
as the negative log-likelihood, to guide the learning process.
5. Applications: Probabilistic models are used in various applications, including regression,
classification, and density estimation. They are particularly useful when it's essential to capture
uncertainty in predictions.

SUPERVISED MACHINE LEARNING: SUPPORT VECTOR MACHINE (SVM)

■ Support Vector Machine (SVM) is a powerful supervised learning algorithm used for both
classification and regression tasks, although it is most commonly used for classification.

■ It works by finding the optimal boundary (or hyper plane) that separates different classes of data
points in a high-dimensional space.
■ Key Concepts of SVM
■ Objective:
– The goal of SVM is to find a hyperplane that maximally separates the classes of data points. In a
two-dimensional space, this hyperplane is a line, but in higher dimensions, it can be a plane or more
complex structure.
– SVM chooses the hyperplane with the maximum margin, which is the distance between the
hyperplane and the nearest data points from each class. These closest data points are called support
vectors.
 9/25/2024

SUPERVISED MACHINE LEARNING: SUPPORT VECTOR MACHINE (SVM)

■ Hyperplane:
– For linearly separable data, SVM looks for a straight hyperplane that divides the classes. The
hyperplane can be represented as:
W*x+b=0
– w is the vector of weights,
– x is the input feature vector,
– b is the bias term.
The model will classify data points based on which side of the hyperplane they fall on.
■ Maximum Margin:
– Margin is the distance between the hyperplane and the closest data points from each class. The
maximum margin hyperplane is the one that maximizes this distance, providing the most robust
classification boundary.
– Support vectors are the points that lie closest to the hyperplane. They play a critical role in
determining the position of the hyperplane.

SUPERVISED MACHINE LEARNING: SUPPORT VECTOR MACHINE (SVM)

■ Non-linearly Separable Data:
– When data is not linearly separable, SVM uses a technique called the kernel trick to map the data
into a higher-dimensional space where a linear hyperplane can separate the classes. This allows
SVM to work for complex datasets with non-linear boundaries.
■ Soft Margin SVM:
– In real-world applications, data is often noisy or may have overlapping classes. Soft margin SVM
allows for some misclassification by introducing a slack variable to tolerate violations of the margin.
This helps to balance the trade-off between maximizing the margin and minimizing classification
errors.
■ Loss Function:
– SVM aims to minimize the following objective function that balances maximizing the margin and
minimizing classification errors:
 9/25/2024

UNSUPERVISED MACHINE LEARNING: K MEANS CLUSTERING

■ A natural approach to similarity based classification is K-Nearest Neighbor (K-NN) which is non-
parametric method used for both classification and regression problems.

■ KNN is one of the most basic yet essential classification algorithms in machine learning. It belongs
to the supervised learning domain and finds intense application in pattern recognition, data mining,
and intrusion detection.

■ It is simple and powerful non-parametric algorithm that predicts the category based on the test
instances according to the k training samples which are closer to the test instances and classifies it to
that category which has the largest probability.

■ There are various distance metrics used in K NN Algorithm they are Eucledian distance, Manhattan
Distance, Minkowski Distance

UNSUPERVISED MACHINE LEARNING: K MEANS CLUSTERING

 9/25/2024

QUESTION BANK

Shiva Prasad KM Assistant Professor CSE Department RYMEC Ballari M: 7899964163

 9/25/2024

MODULE 2

STAFF IN CHARGE
DR. SHIVA PRASAD KM
ASSOCIATE PROFESSOR
DEPARTMENT OF CSE
RYMEC BALLARI

OVERVIEW- IOT SENSING AND ACTUATION

Dr. Shiva Prasad KM Associate Professor CSE Department RYMEC Ballari M: 7899964163