AI encompasses techniques enabling computers to mimic human intelligence for tasks like

reasoning, learning, and decision-making.

Or

building smart machines that can perform tasks that typically require human intelligence.

Artificial Intelligence (AI) - The Broad Goal

 What it is: AI is the overarching field of computer science dedicated to creating

machines that can perform tasks that typically require human intelligence.

 Goal: To make machines "smart" – capable of reasoning, learning, problem-solving,

understanding language, perceiving, and even creating.

 How it works: AI can be achieved through various methods. It can involve:

o Rule-based systems: Explicitly programming machines with "if-then" rules

to handle specific situations (e.g., an expert system for medical diagnosis).

o Search algorithms: Finding optimal solutions by exploring a space of

possibilities (e.g., in game playing like chess).

o Machine Learning: Which we'll discuss next!

 Examples: Virtual assistants (Siri, Alexa), self-driving cars (the entire system, not
just the learning part), expert systems, robotics.
2. Machine Learning (ML) - Learning from Data

 What it is: ML is a subset of AI that focuses on enabling machines to learn from

data without being explicitly programmed for every possible scenario.

 Goal: To train algorithms to identify patterns, make predictions, or take decisions

based on data, and to improve their performance over time as they get more data
(experience).

 How it works: ML algorithms build a mathematical model based on sample data,

known as "training data," in order to make predictions or decisions without being
explicitly programmed to perform the task.

o Supervised Learning: Learning from labeled data (input-output pairs). E.g.,

training a model with images of cats and dogs explicitly labeled as such, so
it can classify new images.

o Unsupervised Learning: Learning from unlabeled data to find hidden

patterns or structures. E.g., clustering customer data into different segments
without prior knowledge of those segments.

o Reinforcement Learning: Learning through trial and error, by interacting

with an environment and receiving rewards or penalties. E.g., training an AI
to play a video game by rewarding it for good moves.

 Examples: Email spam filters, recommendation systems (Netflix, Amazon), fraud

detection, predictive analytics.

3. Deep Learning (DL) - Learning with Neural Networks

 What it is: DL is a subset of Machine Learning that uses Artificial Neural Networks
(ANNs) with many layers (hence "deep") to learn complex patterns and
representations from data.

 Goal: To enable machines to learn from vast amounts of complex, often

unstructured data (like images, audio, text) by mimicking the way the human brain
processes information. DL models can automatically extract features from raw
data, eliminating the need for manual feature engineering (a common step in
traditional ML).

 How it works: DL relies on deep neural networks, which are inspired by the
structure and function of the human brain. These networks consist of multiple
layers of interconnected "neurons" that process information hierarchically. The
"deepness" refers to the number of hidden layers between the input and output
layers.
 o Each layer learns increasingly complex features from the raw input. For
example, in image recognition, early layers might detect edges, middle
layers might detect shapes, and deeper layers might identify complete
objects.

 Examples: Facial recognition, natural language processing (understanding and

generating human language, like in chatbots and translation tools), speech
recognition (Siri, Google Assistant), self-driving cars (specifically, the perception
and decision-making components based on visual input), medical image analysis.

Analogy:

Imagine you want to teach a computer to identify different types of fruits.

 AI (the big goal): The overall aspiration to make a computer that can identify fruits
as well as a human.

 ML (one way to achieve it): Instead of writing explicit rules for every fruit (e.g., "if
red and round and small, it's a cherry"), you feed the computer thousands of fruit
images with their labels (apple, banana, cherry). The ML algorithm learns the
patterns and distinguishes them on its own.

 DL (a powerful type of ML): You use a deep neural network to process the fruit
images. This network can automatically learn features like color, shape, texture,
and even subtle nuances, without you having to tell it what features to look for. It's
particularly good at handling complex, raw visual data and achieving very high
accuracy.

Examples • Virtual Assistants: Siri processes voice commands to perform tasks. • Game
Playing: AlphaGo uses AI to strategize in the game of Go. • Recommendation Systems:
Netflix suggests content based on user preferences.

AI = ML + DL + Rule-based Systems + Search + Planning + Logic + Reasoning, etc.

This course is about

Deep learning, a subset of machine learning, has gained prominence over the past decade
due to its effectiveness in various domains. Applications include speech recognition,
computer vision, and natural language processing. Stateof-the-art systems from major
technology companies rely on deep learning as a core technology.

2 Course Structure The course covers foundational concepts and building blocks of deep
learning, progressing from basic to advanced topics
2.1 Fundamental Building Blocks • Single Neuron: Introduction begins with the perceptron
and sigmoid neuron, explaining their role as basic computational units.

• Multilayer Perceptrons (MLPs): Neural networks comprising multiple layers of neurons,

known as multilayer perceptrons, are explored.

• Training Algorithms: Backpropagation, utilizing gradient descent, is introduced as the

primary method for training neural networks.

• Applications: Feedforward neural networks are applied to tasks such as autoencoders and
other architectures.

2.2 Recurrent Neural Networks (RNNs) Recurrent neural networks address sequential data,
prevalent in domains like natural language processing and speech recognition. Sequences
appear in:

• Sentences as sequences of words or characters.

• Speech as sequences of phonemes.

1 • Videos as sequences of images.

Tasks involving sequential data include classification and sequence prediction, such as
translating a sentence from a source language to a target language. Training RNNs involves
backpropagation through time, a variant of backpropagation. Challenges in training RNNs
are addressed by advanced architectures like Long Short-Term Memory (LSTM) units and
Gated Recurrent Units (GRUs).

2.3 Convolutional Neural Networks (CNNs) Convolutional neural networks primarily apply to
computer vision tasks. CNNs use convolutional operations to derive hierarchical, abstract
representations of images for tasks such as: • Classification • Object detection • Image
segmentation The structure and operation of CNNs are compared to feedforward neural
networks to highlight differences.

2.4 Encoder-Decoder Models

Encoder-decoder models integrate feedforward neural networks, RNNs, and CNNs to process
various inputs (e.g., images, speech, text) and generate outputs.

Applications include:

• Image captioning: Encoding an image to generate a textual description.

• Machine translation: Converting a sequence in one language to another.

• Document summarization: Condensing text into concise summaries.

2.5 Attention Mechanism The attention mechanism enhances encoder-decoder models by

focusing on critical input components. For example: • In image captioning, attention
prioritizes key objects (e.g., a boy, frisbee, and park) over background elements. • In
document classification, attention identifies important words indicating the document’s
category (e.g., politics, sports, finance).

Dendrites:Inputs

Cell nuclues(Soma):node

Synapse:weights

Axon:output
Circle at the end indicate inhibitory input if the inhibitory input is 1 the output will be 0

Geometric interpretation of McCullah Pitts unit?