Agenda

  • What is AI
  • Building Blocks of AI
  • History of AI
  • Machine Learning
  • Types of Machine Learning
  • Big Data
  • Use Cases of AI and ML
  • Demo of a AI use case

What AI brings to mind...

Self Driving Cars
Credit Card Fraud
Face Recognition
Self-Driving Cars Credit Card Fraud Detection Face Recognition
Early Detection of Terminal Illnesses
Smart Voice Assistant
Robots
Early Detection of Terminal Illnesses Smart Voice Assistant Robots
https://youtu.be/GCnCLgrTkec

Wait..What?

AI on the cloud?

Deep Learning?

Watson?

Classifier?


Don't worry..We got you covered

For so long, Artificial Intelligence (AI) has been just a tool that helps solving simple problems and performing specific tasks.

This was achieved through programming and declaring the steps a computer should follow to achieve the task.

But lately, the concepts of Machine Learning and Deep Learning are starting to show so much potential, and are promising to disrupt many industries and change the future.

AI Landscape

Artificial Intelligence Icon Artificial Intelligence
Machine Learning Icon Machine Learning
Deep Learning Icon Deep Learning
AI is the umbrella term that includes other subsets. It involves explicitly programming a computer to perform a specific task. It is used to help solve simple problems.
Machine Learning is a concept that entails a program that learns without being explicitly programmed. It basically learns from data and experience.
Deep Learning utilizes Neural Networks to understand and learn from data.
https://community.hpe.com/t5/Behind-the-scenes-Labs/Labs-Deep-Learning-Cookbook-headlines-the-launch-of-HPE-s-AI/ba-p/6981300

Evolution of AI over the last decades

AI Timeline

http://bisintek.com/science/2017/12/27/knowing-basic-artificial-intelligence/

Let's simplify this even more

Machine Learning and Deep Learning borrow a lot from biology, how our brains are wired and the way we learn as humans.

So let's take a quick look on that first.

As humans, we are born with a blank slate (somewhat).

We learn through input data we receive from our environment and our senses.

Brain Learning Process

We incorporate this input data with the outcome to form experiences and store them in our memory.

Brain Learning Process

Experiences that are repeated and accompanied by some form of reward or punishment create a neural pathway in our brains.

Brain Learning Process

Neural Pathways make it easier for the brain to retreive past experiences in new, similar situations and use what it previously learned.

Brain Learning Process

https://www.youtube.com/watch?v=ELpfYCZa87g

We found a way to make machines learn the same way...

Machine Learning

+ Experience
Data  Computer Performing a Task Task  Performance 

What does a Machine do with the Data?

And how does it actually learn?
A Machine tries to find a relationship from the input (X) to the ouput (Y) Finding a Model
This process is called developing a Model* from the data, and it is up to the Machine Learning Algorithm* to find that Model

*Model: is a representation of a real-world process. It helps explain a system, study the effects of different components, and to make predictions about future behaviour

*Algorithm: a set of steps to follow by a program to solve a problem

https://www.cs.toronto.edu/~frossard/post/tensorflow/

How does the Machine learn to get the best Model?

Through configurable parameters

Let's take an example

Below is a simple Linear Regression Linear Regression
This model can be represented as

Y = aX + b

where "a" is the slope and "b" is the intercept.

Both "a" and "b" are Configurable Parameters that the learning algorithm can tune to find the best Model

How does a Learning Algorithm know it is correct?

We measure its Performance by calculating a Cost Function Measure Performance

Simply put, it is a way to estimate how wrong is the Model we have now in terms of estimating the relationship between the input (X) and the output (Y)

The Learning Algorithm's mission is to find Parameters that minimize the cost function

http://bestanimations.com/Science/Science.html
Model Training

https://medium.com/onfido-tech/machine-learning-101-be2e0a86c96a

This process is called Model Training

By the end of this process, we have a model that describes relationships in our data, now we can use that model to predict an output (that we don't know) from new input (future data)


That's the essence of AI

What problems can Machine Learning solve?

Spam Filter

Spam Filteration

Product Recommendation

Product Recommendation


Medical Diagnosis

Medical Diagnosis

Targeted Marketing

Targeted Marketing

So Machine Learning is only able to solve very simple problems..


What about the more complex tasks like

  • Machine Vision,

  • Natural Language Understanding



and all those things that make AI so cool?



It needs advanced feature engineering skills for such problems..is there an alternative?

Let's take this to another level..


Deep learning

Deep Learning

+ Experience
Data  Computer Performing a Task Task  Performance 
http://www.china-vision.org/user/FLIR%20(%E5%8E%9F%20Point%20Grey)/plan_detail/26982.html

Let's Take an Example

Convolutional Neural Network

https://becominghuman.ai/building-an-image-classifier-using-deep-learning-in-python-totally-from-a-beginners-perspective-be8dbaf22dd8

Building block of a Neural Network

Artificial Neuron

https://harishnarayanan.org/writing/artistic-style-transfer/

Comparing Artificial Neuron to a Human Neuron

Human Neuron

https://medium.com/typeme/lets-code-a-neural-network-from-scratch-part-1-24f0a30d7d62

Stacking Many Neurons Together


Humans Human Brain

~100-1,000 Trillion synapses

Deep Neural Network Deep Neural Network

~1-10 Billion Artificial Neurons

Humans have ~10,000 times more computational power than the most powerful "computer brains"

http://neuron.com.au/ https://www.reddit.com/r/gifs/comments/6elloq/neural_network_3d_simulation/

What's special about Deep Learning?

It can automatically detect millions of features that can be very subtle, but combined, they can be the bases of a machine's understanding of unstructured data like Images and Sound
Machine Learning vs. Deep Learning

Ok, but how does a Neural Network actually learn?

https://medium.com/swlh/ill-tell-you-why-deep-learning-is-so-popular-and-in-demand-5aca72628780

Backpropagation

Backpropagation

http://www.fictiontofact.com/technology/ai/ai-detectives-are-cracking-open-the-black-box-of-deep-learning/

What Problems Can be Solved with ML and DL?


Anything that falls under the following categories

Machine Learning Types

https://nowenlightenme.com/2018/03/18/types-of-machine-learning/

Supervised Learning

Data + Labels (What this data means)


Classification
Categorical Output Classification

Examples:

  • Spam Filteration
  • Image Classification

Regression
Continuous Output Regression

Examples:

  • Predicting Stock Price
  • Predicting Arrival Times in Navigation Apps

https://www.pyimagesearch.com/deep-learning-computer-vision-python-book/ https://www.cs.toronto.edu/~frossard/post/tensorflow/

Unsupervised Learning

Data + No Labels


Clustering Clustering

Examples:

  • Customer Segmentation
  • Targeted Marketing

Anomaly Detection Anomaly Detection

Examples:

  • Unusual Behaviour in surveillance cameras
  • Sensor faults or unusual signals

http://www.neos.hr/tableau/what-is-it/ https://unsupervisedlearning.wordpress.com/2014/08/04/topological-anomaly-detection/

Reinforcement Learning

Learning Algorithm has memory. Learns from Experience & Mistakes


Reinforcement Learning

https://simple.wikipedia.org/wiki/Reinforcement_learning

This is now..

Smart Voice Assistants
Robot Playing Chess
Robot Playing Football
Self Driving Car

Imagine tomorrow's possibilities..

Can we reach General Artificial Intelligence in the near future..?



Thank You!

https://youtu.be/_Xcmh1LQB9I