Data,

Data and
More Data!

Presented by-

Samriddhi Guha (UEMK 3rd Year, CSE)

Mayukh Ghosh(UEMK 1st Year, CST)
Sagnik Mishra(UEMK 1stYear, CST)
Yash Vrat Shrivastava(UEMK 1st Year, CSIT)
st
Artificial Intelligence (AI)
Rise of AI

- The term AI was coined in the year 1952 by John McCarthy.

- IBM Deep Blue defeated Int’l Chess Player Garry Kasporov(1990s)
- Development of Driverless Cars, assistants like Siri/Google Assistant.(2010s). This period was referred to as Connectionist.
- Models of GPT and Dall-E were brought in 2018 and 2021 respectively.(These are narrow Ais answering questions and
solving tasks to a certain extent).
- Rise of Deep Learning in 2020s.

Implementing AI
We implement AI by the concept of Machine Learning, i.e artificially designing algorithms to give predictions based on the
environment and certain factors.
 How does Machine Learning work?
A Machine Learning system learns from historical
data, builds the prediction models, and whenever it
receives new data, predicts the output for it.

Need of ML

- Machine learning is the study of computer algorithms that allow computer

programs to automatically improve through experience
- Broadly divided into 3 categories– Supervised ,Unsupervised (Group and
interpret data based only on input data) and Reinforced learning
Data Oriented Career Paths

DATA ANALYST DATA SCIENTIST

-Have advanced
-Little to no degree in
experience / computer science DATA ENGINEER ML ENGINEER
background to -knowledge of
invent physics , statistics -Builds a data -Manage the
new algorithms and applied pipeline to clean, implementation
-Good mathematics transform and -Manage the data
understanding on -Knowledge of organise messy pipeline that goes
existing tools for inventing new data into resources into the model and
problem solving algorithms to solve for being used by a the decision
-Understanding of data problems data scientist / pipeline coming
statistics -Ability to deliver analyst out of the model.
programming, recommendations
machine learning,
data visualization
Domain Expertise
-Strategy Planning and Execution
-Business rules and policies
-Regulations and Compliance
-Types of Marketing
-Accounting Standards
-Supply Chain and Logistics
-Consumer Psychology
-Ethics in Applications

People Skills

-Empathy
-Integrity
-Communication Skills
-Leadership Skills
-Human Behaviour and Body Language
Deep Learning

What is Deep Learning?

Deep learning is a subset of machine learning
where neural networks with multiple layers (deep
architectures) learn complex patterns from data,
enabling tasks like image recognition and natural
language understanding.

It combines the features of big data and machine

learning,
removing the problem of feature selection and
extraction in ML.

Although it is effective, it lacks human touch,

making it feel robotic and lack empathy.
Deep Learning

What is Deep Learning?

ANN(Artificial Neural Networks) try

to replicate the working of the
human mind while decision making.

This is an MLFFNN(Multi Layer

Feed Forward Neural Network,
used to solve everyday life complex
problem)
Urban Analytics and Social Sensing

An Introduction

-Urban analytics is the practice of using new forms of data in combination with computational approaches to gain
insight into urban processes. Increasing data availability allows us to ask new and often complex questions about
cities their economy how they relate to the local and global environment and much more.

-The term social sensing describes crowd-sourcing techniques and applications that make use of sensors that are
closely attached to humans (e.g., as parts of smartphones) and are either directly or indirectly used to provide
sensor observations at a high spatial and temporal resolution.
Urban Analytics and Social Sensing

Scope of Urban Analytics

-Used to solve problems via Social Media Analytics, Community Detection, Online Controlled Experiments,
Education and Occupation.
-The broad uses for it are in solving the problems of determining POI(Places of Interest), trajectories and suitable
recommendations to suit people.

Scope of Social Sensing

-Data from social media apps like Twitter help us in monitoring research diffusion in a global pandemic like
COVID-19 by the use of social-spatial sensors.
-Better understanding of places and sentiments.(Positive, Negative and Sentient levels)
-Spatio-Temporal Event Detection
Urban Analytics and Social Sensing

Recommendations
-Recommendations utilise the concepts of both social sensing and urban analytics to make meaningful suggestions.
-Applications include queue aware Personal Tour Recommendation
-Tour Recommendations with Recency Updates
-Group Tour Recommendations.

Major Problem Statements

The majorly discussed research questions are as follows:

-Sentiments and emotions of people during the COVID-19 pandemic.

-Change in sentiments and emotions w.r.t global events.
-Common topic of interest during the pandemic.
Urban Analytics and Social Sensing

Approach of Solution
The algorithm consists of the following steps:

- Building of Quad Tree by using a rectangular decomposition of space with minimal size.
- Event Detection using a Poisson Distribution
- Merging of Events in close time proximity
- Pruning of events below a specific durations

- Examples:
- QueueTourRec (A queue sensitive solution for reducing waiting times by assessing the queues at regular time
intervals)
- PersTour (Selection of POIs for individuals based on their tastes)
- GroupTourRec (Selection of like-minded individuals, their POIs and a suitable tour guide)
Neural Networks

Data Storage and Big Data

-Data storage is a big problem in modern day, as data is needed for understanding the context, in order to make more sentient and
better performing AIs.
-We are concerned with Volume(scale of data), Velocity(data streaming), Variety(different forms), Veracity(uncertainty of data).
-Big Data is dealt with in a discipline of ML known as Deep Learning.
-AI has impacted the job sector severely, with AI replacing many entry level jobs.

Neural Networks

NNs mimic the functionality of the human brain, by the use of Perceptrons(Artificially made Neurons). These help take important
decisions and solve everyday complex programs with the help of matrix form. There are two types of NNs presently in use -
RNN(Recurrent NN) and CNN(Convolutional NN).
Earlier, Feed Forward NNs were used, but quickly discarded due to vanishing gradients, unstable weights, etc.
Nowadays, word embedded systems are used.
Types of Word Embeddings
-Word2Vec
-GLoVE
-FastText
-ELMo
-BERT
Transformers
The innovation behind Transformers boils down to three main concepts
-Positional Encodings
-Attention
-Self-Attention

They are a concept by which we encode words into vectors for the
machines to understand and interpret them whie developing LLMs.

Recurrent neural nets

-CNNs are FFNNs : they cannot handle sequences well
-We need a temporal model for learning long sequences of
words , e.g., temporal changes in long product reviews
Factors to consider while developing an AI Machine
-How would they think?
-How would they look?
-What range of emotions would they display?
-How smart should they be?
-How fair would they be?

But, could they ever be really fair in a world of such complex everyday scenarios?
This gives rise to the fatal risk of humans wiped out due to fairness in the eyes of the AIs, among
many others.

Hence, we should design AIs on the basis of the following seven pillars:-

-Multidisciplinarity
-Task Decomposition
-Parallel Analogy
-Symbol Grounding
-Similarity Measure
-Intention Awareness
-Trustworthiness
Generative AI:
-Generative AI is a form of Artificial Intelligence that generates multimedia(audio, video, images) on
the basis of prompts.
-Y=f(X), where Y=output and x is the prompt. For eg, speech to text converion, where audioclips are
taken as input to return text outputs.

Types of ML Functions:
-Regression
-Classification
Using this we model predictive models known as Large Language Models. These are a part of
GAIs(Generative AIs that generate a specific object, with infinitely many possibilities )
LLMs are designed as follows :-

-Pre Training (Knowledge of Language, World and Semantics)

-Fine Tuning (Final changes by the help of human aided changes)
-Reinforcement by RLHF/RLAIF(Reinforcement learning from Human
Feedback or Reinforcement Learning from AI Feedback).

In conclusion, LLMs are designed to create many libraries and APIs to

further simplify the designing of chatbots, GPTs, etc.
Data Science in the age of Generative AI:
-Data Science is the study of numbers to find out patterns in it, whereas generative AI works on an
input to generate desired output based on a certain prompt. As we combine both, we get the domain
of RAG(Retrieval-Augmented Generation). RAG combines the knowledge of LLMs and Data Science
to generate New Languages.
Steps in implementing RAG:
-Build knowledge base
-Divide into small facts
-Convert the files into vector representations
-Vectorial embedding of questions
-Every vector is assigned some score, and the vector with score closest to that of the sentence is returned.
-Make an API to call OpenAI with context for the answer
-Perform post-processing and send back

LLMs based solutions are approached as follows :-

-Formation of hypothesis
-Collection of Data
-Insights Gathering
-Accept/Reject the Hypothesis based on the collected insights
The Trap of Data Science
-It may so happen that a successfully trained model fails to perform in a real life environment after
deployment, then we call it failed. To amend these problems we check the correlation between
x(features) and y(the corresponding output).
Correlation vs Causality

Correlation: refers to a statistical relationship between two variables. When two variables are correlated,
changes in one tend to be associated with changes in the other. However, correlation alone does not imply
causation.

Causality, on the other hand, indicates a cause-and-effect relationship. It suggests that changes in one
variable directly lead to changes in another. Establishing causality often requires rigorous experimentation
or well-designed observational studies.
Given enough data, two completely unrelated things can show a correlation.
Example 1:

Example 2 :
Rate of kidnapping in summers showed a positive correlation
with rate of ice cream sold in summers in an observation
sample area. However, this is purely coincidental as ice cream
is sold more in summers , and more people loiter outside more
making them easy target for criminals
 RPA-Definition

RPA (Robotic Process Automation) helps make mundane and repetitive tasks easy and able to be performed by
the bots using automation. Once designed, these automation robots do not need human intervention and can
work for hours on an end, to simplify repetitive tasks that consume a lot of time and efforts.

It enables cost cutting and saves a lot of money and resources in a single fiscal year.

Software Used for RPA: UiPAth Studio

url: https://cloud.uipath.com
Hands on approach on RPA:
 Introduction to IoT(Internet of Things)

-The IoT (Internet of Things) connects physical objects to the Internet– like common household objects
like bulbs, wearables, etc.

-The major types of devices are switches(for sending a command) or sensors(that collect data for
sending elsewhere)

-Modern day IoT comprises- Sensors + Internet + Big Data + Artificial Intelligence, which take data inputs
via sensors and then process it to get the data required for performing tasks. This combination is also
referred to a AIoT(Artificial Internet of Things).

Applications:
- Cost cutting for enterprises(we can purchase once and deploy globally)
-To create more IP addresses, as we are running out of 16 bit Ipv4 addresses.

Goals:
-Cost Leadership
-Industry Leadership
 Introduction to CyberSecurity

Cybersecurity is a domain of computer science wherein the data we use and store everyday is protected.
Nowadays, a lot of our sensitive data, for eg: Social Security numbers, Bank Details etc have to be put
out on the internet for the fulfilment of multiple tasks.

The changing shifts in CyberSecurity

Earlier considered to be a topic of interest for Computer Science only, now CyberSecurity has gained
importance outside of Computer Science to gain the status of Behavioural Science due to the growth in
the requirement of storing and protecting data. Companies rely heavily on IoT based solutions for scaling
up their businesses.

CyberSecurity Life Cycle

-Find Position
-Know Position
-Create Defense Layer
-Create Predictive Layer
Scope of AI

Artificial Intelligence
The over reaching concept of machines capable of mimicking or augmenting human intelligence.
AI systems can perform tasks traditionally requiring human cognitive abilities, including problem-
Artificial solving, recognizing patterns, and understanding language .
Intelligence
Machine Learning / Deep Learning
Machine / Deep
Learning These are Al systems that learn from experience. They refine their performance over time by
processing more data and learning from the patterns they observe.
Uses artificial neural networks with multiple layers, or 'deep' structures, to model and understand
Generative AI complex patterns.

Large Language Gen Al

Models These models can generate new, unique content after learning patterns from their
training data. Whether it's producing a new piece of music or proposing an innovative
product design, Generative Al takes creation to a new level in the world of Al.

Large Language Models

Artificial neural networks that can generate and understand natural language text and other
types of content, by learning from massive amounts of data and using attention mechanisms
and transformer architectures
 Hands on Sentiment Analysis App

A Sentiment Analysis App is a usage of LLM(Large Language Model) for understanding the emotions of a
speaker while in a conversation. It is helpful in many scenarios like in understanding human better emotions via
chats to help

Software Used for Sentiment Analysis App: Anaconda, Jupyter Notebook for running Python(version 3.7 and
later)
Modules Required: Streamlit (for UI), OpenAI (for accessing the LLM).

It can successfully predict the moods of the person by taking a text input from the person.