Headstart in Deep-Tech

Microsoft AI Workshop

Credits: Microsoft

What is Artificial Intelligence?

Theory and development of computer systems able to perform tasks normally requiring Human Intelligence.

Examples: Visual perception, Speech recognition, Decision making, Translation between languages, etc.

History of Artificial Intelligence in a glimpse.

1950 — “Turing Test”

A Turing Test is a method of inquiry in artificial intelligence (AI) for determining whether or not a computer is capable of thinking like a human being.

1956 — “Dartmouth workshop”

Birth of Artificial Intelligence.

1966 — “ELIZA” passed the “Turing Test”.

Credits: Steemit

Eliza was the first attempt to create a natural language processing computer program from 1964 to 1966 at the MIT Artificial Intelligence Laboratory.

1970 — “AI Winter”- Pessimism about the Machine Learning effectiveness.

1990 — Machine Learning was recognized as a separate field that started to bloom in the industry.

2010 — Machine Learning becomes integral to many widely used software services.

What should be done by a Data Scientist?

1. Sharpen the ability to handle DATA.
2. Focus on the TECHNIQUE, not the TOOL.
3. Increase your application of TECHNIQUES.
4. Pick up a strong Secondary skill and collaborate with the team.
   i) Domain
   ii) Project Management
   iii) Information design etc.

STAND OUT OF THE CROWD

Credits: Microsoft

Types of Learning

1. Supervised learning
2. Unsupervised learning
3. Semi-supervised learning

Algorithm

1. Regression
2. Classification
3. Clustering
4. Neural Network

Regression

Derived from statistics. Used widely in Machine Learning,

Linear Regression

Main Goal: Draw a straight line that best fits all the data points on the scatter plot. The best fit line is also known as the regression line.
The regression line should have a minimum value.
First-degree cure: y=mx+c

Advantages:
1. Simple to understand.
2. Based on mathematics.
3. Usually the training time taken is low.
4. Continuous value prediction.

Challenges:
1. Too simple for a lot of real-world problems.
2. Cannot handle it if it is more complicated.

Modern Application
1. Predicting demand.
2. Stock prediction.
3. Credit assessment etc.

Classification

Predicts discrete value. i.e. (0 or 1) or (Yes or No). Classification issued for categorical response value, where the data can be separated into specific “classes”.

Two-class classification

Predicts between two categories. Yes or No
Examples:
Is this tweet true?
Will the customer renew their existing service?

Multi-class classification
Predicts between several categories.
Examples:
What is the mood of the tweet?
Which service will the customer choose this time?

Decision Tree

Credits: kdnuggets

The decision tree is a flowchart like structure created upside down with the root on the top, internal nodes contain conditional based on which trees is split into branches. Leaf nodes(terminal node) holds a class label.
Example: Will the kids play cricket today?

Key idea
Look at the past data. Determine the attribute that best classifies the past data. Use that attribute at the root of the tree.
Repeat this process at each branch.

Advantages
1. White box, easy to interpret and explain.
2. Used in finding the most important attribute.
3. They are not affected by outliers, less data cleaning is required.
4. Once created, this can provide fast classification.

Challenges
1. Requires a lot of data.
2. Not good at regression.
3. Limited to one output without a problem.
4. Works better with discrete value attributes compared to continuous value attributes.

Modern Application
Used in the field of Astronomy, financial analysis, power systems, molecular biology, etc.

Clustering — K means

Credits: geeksforgeeks

Clustering is a method of unsupervised learning and a common technique for statistical data analysis used in many fields.
K signifies the number of clusters.

Advantages
1. No need for classified input data.
2. No need information about attributes.

Challenges

1. Results might differ with different successive runs.
2. Need to specify the number of clusters.

Modern Application
Used widely in the domain of e-commerce, credits score, supermarket, document classification, etc.

Neural Network

This is similar to a biological neuron. This is also termed as an Artificial Neural Network(ANN).

Advantages
1. It goes well with continuous value attributes
2. Can have a lot of attributes.

Challenges
1. Blackbox results. Not able to check or understand the results.
2. Training takes a lot of time.

Modern Application
Used for predictive analysis, fraud detection, any kind of cognitive skills, etc.

Responsible Artificial Intelligence

Principles

1. Fairness
2. Reliability and safety
3. Privacy and security
4. Transparency
5. Accountability
6. Inclusiveness

Challenges or Risks

1. Bias can affect the system.
2. Errors can cause harm.
3. There are chances for our data to be exposed.
4. Users must trust a complex system.

STAY CURIOUS AND STAY RELEVANT