Some ideas on Artificial Intelligence, Machine Learning, Neural Networks

1. Artificial Intelligence
Conceptual Overview

2. Introduction to A.I.
Content Overview
1. Artificial Intelligence
2. Neural Networks
3. Machine Learning

3. What is A.I.?
Artificial intelligence (AI) is intelligence exhibited by machines.
Every computer program can be classified as artificial intelligence.
As machines become increasingly capable, mental facilities once thought to require intelligence are removed from the definition. For instance,
optical character recognition (OCR) is no longer perceived as an example of "artificial intelligence", having become a routine technology.
ABOUT A.I.
A revolution is heading our way

4. Capabilities currently
classified as A.I.:
01
Human Speech
Successfully
understanding human
speech competing at a
high level in strategic
game systems (such as
chess and Go)
02 03 04 05
Autonomous Cars
Cars that are guided by
trained Artificial
Intelligence
Routing
Intelligent routing in
content delivery
networks
Military Simulations
Military simulations that
require trained models to
create a comprehensive
solution
Complex Data
Interpreting complex
data using trained
models

5. What are Neural Networks
Biological neuron: Dendrite (INPUT), Axon ( PROCESS), Axon Terminal (OUTPUT)
Node: Input Parameter (INPUT), Function (PROCESS), Output Parameter (OUTPUT)

6. What are Neural Networks
Artificial neural networks (ANNs) or connectionist systems are computing systems inspired by the biological neural networks that constitute animal brains. Such systems learn
(progressively improve performance) to do tasks by considering examples, generally without task-specific programming.
Neural networks are a set of algorithms, modeled loosely after the human brain, that are designed to recognize patterns. They interpret sensory data through a kind of machine
perception, labeling or clustering raw input.

7. What are Neural Networks
Deep artificial neural networks refers to the number of layers in a neural network. A shallow network has one so-called hidden layer, and a deep network has more than one.
Multiple hidden layers allow deep neural networks to learn features of the data in a hierarchy, because simple features (e.g. SUM, COUNT) recombine from one layer to the next,
to form more complex features (e.g. Determine if it is time to pick up your hand).

8. What is Machine Learning
The “learning” part of machine learning means that ML algorithms attempt to optimize along a certain dimension; i.e. they usually try to minimize error or maximize the
likelihood of their predictions being true. They are, in short, an optimization algorithm. If you tune them right, they minimize their error by guessing and guessing and guessing
again.
ML, using Neural Networks help us cluster and classify. You can think of them as a clustering and classification layer on top of data you store and manage. They help to group
unlabeled data according to similarities among the example inputs, and they classify data when they have a labeled dataset to train on.

9. How can A.I. be used?
The greatest innovation since the inception of the internet
Clustering Classification Predictive Analytics

10. Classification
Detect faces, identify people in images, recognize facial
expressions (angry, joyful)
Identify objects in images (stop signs, pedestrians, lane
markers…)
Recognize gestures in video
Detect voices, identify speakers, transcribe speech to text,
recognize sentiment in voices
Classify text as spam (in emails), or fraudulent (in insurance
claims); recognize sentiment in text (customer feedback)

11. Clustering
Clustering or grouping is the detection of similarities. Deep
learning does not require labels to detect similarities. Learning
without labels is called unsupervised learning.
Search: Comparing documents, images or sounds to surface
similar items.
Anomaly detection: The flipside of detecting similarities is
detecting anomalies, or unusual behavior. In many cases,
unusual behavior correlates highly with things you want to
detect and prevent, such as fraud.

12. Predictive
Analytics
Hardware breakdowns (data centers, manufacturing, transport)
Health breakdowns (strokes, heart attacks based on vital stats
and data from wearables)
Customer churn (predicting the likelihood that a customer will
leave, based on web activity and metadata)
Employee turnover (ditto, but for employees)
Stocks trends

13. Relevance to
Enterprises
AML Pattern Detection & Fraud detection
Chatbots (Customer support)
Customer recommendations (Products, Packages)
Sentiment analysis (Voice of the customer)
Credit score
Marketing classifications
Algorithmic trading

14. How is A.I. implemented?
01
Problem
Identification
02 03 04 05
Data
Gathering
Building and
training the model
Testing
the model
Run in
Production

15. Problem
Identification
How is A.I. Implemented
01
Is my problem supervised or
unsupervised?
02
If supervised, how many labels am I
dealing with?
03
What’s my batch size?
The final purpose it is to identify what is the best optimisation algorithm and the best technical approach for solving the problem

16. Data Gathering
How is A.I. Implemented
What would be the data structure ?
Define example data / training set. Min 50.000 records.
Define test data. A set of example data which was not included in previous training set.
The training set is used to teach, the test set is used to check
In the machine learning world, this is called supervised learning, because we have the correct answers for the images we’re making guesses
about.

17. Building and Training the Model
How is A.I. Implemented
As soon as the problem is completely identified and example/test data is prepared, it must be decided which optimization algorithm can be used
in order to solve the problem. The following table attempts to show the neural nets most useful for different problems:
Data Sector Use Case Input Transform Neural Network
Text Sentiment analysis Word Vector Gaussian Rectified RNTN or DBN
Document Document classification Word Count Probability Binary DBN or Stacked Denoising Autoencoder
Image Image Search Gaussian Rectified Deep Autoencoder
Sound Voice Recognition Gaussian Rectified Recurrent Net
Time Series Predictive Analytics Gaussian Rectified Recurrent Net

18. Building and Training the Model
How is A.I. Implemented
The actual training consists of loading all example data into the model, and allow the system/program to identify the best applicable function
able to transform all provided input data into the provided output data.
This process is done by guessing, adjusting, guessing again until the most optimum function is found with the minimum error margin. This
process requires intensive calculation with big amounts of data.
Each step for a neural network involves a guess, an error measurement and a slight update in its weights, an incremental adjustment to the
coefficients.
It has to start out with a guess, and then try to make better guesses sequentially as it learns from its mistakes. A neural network is a corrective
feedback loop, rewarding weights that support its correct guesses, and punishing weights that lead it to err.

19. Testing the Model
How is A.I. Implemented
The actual testing consists of running the test data through the newly defined and trained model, with the purpose of identifying what is the
accuracy of the model.
There is no true or false in A.I. All results will be expressed in percentage values, without having 100% or 0% as a result. Still, 90% accuracy is
considered a very bad result for A.I. model.
Eg:
a) INPUT: { “name” : “Osama bin Laden”, “dateOfBirth” : “10 March 1957”}, OUTPUT: “TERRORIST” 99.52%
b) INPUT: { “name” : “Alin IFTEMI”, “dateOfBirth” : “11 October 1980”}, OUTPUT: “PROGRAMMER” 99.27%

20. Run in Production
How is A.I. Implemented
As soon as the model is well defined, trained and tested, it can be run and executed in production.
It is important to understand that from time to time, each model needs to be retrained and updated with new changes, even if it is still valid and
works reasonably good.
Main reason for this operation:
- Data is in continuous change
- New classification labels might appear
- The system needs to be updated with confirmed labels all the time
Still, these operations are not required to be executed very often.

21. Great Implementations
- Siri / Alexa
- Tesla
- Amazon (predictive orders)
- Netflix (suggestions)
- AlphaGo / DeepMind

22. Blockchain Implementations ?
- Consensus
- Estimate transactions no in private networks, to ensure optimal hardware power
- Help AI explaining itself
- Increase artificial trust

23. Thank you
Follow us on
https://modex.tech/
https://twitter.com/modex_tech
https://t.me/joinchat/GNE4JQxWumaWJxaSn3ksMw
https://www.facebook.com/modex.tech.team
https://www.instagram.com/modex.technologies/