Key technological concepts for AI & Data

Analytics Translator

An analytics translator is someone who bridges the gap between the technical world of data engineers, analysts and scientists with the operational business-focused marketing, supply chain, maintenance, risk management e.t.c

Analytics translators perform the task of:

• Communication bridge: facilitate effective communication between technical and business teams
• Understanding business objectives: they deeply understand the business goals, challenges and requirements of departments within an organization.
• Data storytelling: they interpret complex data analysis insights in a way that resonates with non-technical stakeholders
• Guiding decision-making: By using analysis insights, they can provide guidance and recommendations to business leaders and operational teams

Data Analytics vs Data Science

Similarities

• They both work with large data sets
• They leverage programming to do their jobs (query explore and manipulate data)

Differences

• Data analysts look for answers (why are our sales reducing at the beginning of the year, how much time are users spending on our products)
• Data scientists look for ideas (imagine we could reduce the amount of time it takes to recommend products accurately to our uses)

Data scientists are much more focused on innovation, and you will find them working with machine learning, deep learning, database management and the building of algorithms. They determine how to build AI & data products.
Data analysts on the other hand determine what AI & Data Products to build and how the products perform

Types of Artificial Intelligence

Artificial Narrow Intelligence (ANI): This is the type of AI that can match human intelligence in one specific domain such as Image recognition or speech recognition

Artificial General Intelligence: This is AI that can match human’s overall intelligence

Artificial Superintelligence: An AI that can exceed human intelligence. this does not exist today but is regularly depicted in science fiction movies

Machine learning (inspired by the way humans learn uses statistics)

This is a subset of AI that enables machines to learn from experience through the use of large data sets, statistics and algorithms. The output of this process is a machine-learning model. Machine learning involves learning through the finding of patterns in a large dataset. For example, a machine will be able to identify a dog. after it has been shown the image of a million dogs. This is also how humans learn.

Deep Learning (inspired by the way the human brain works uses neural networks)

This is a subset of machine learning. it’s a much newer method of building artificial intelligence. Most of the advancements in image recognition and speech recognition have been achieved because of deep learning.

Deep learning mimics the human brain by using what we call artificial neural networks (ANNs). There are two types of Neural networks:

Convolutional neural networks (CNN) — used in image recognition

Recurrent neural networks (RNN) — Used in speech recognition

Deep learning vs machine learning

Deep learning in theory can achieve the best overall performance for AI.

• Deep learning is more technically complex and also more expensive to run
• Machine learning models can run on your home computer hardware, but deep learning models require GPUs for computation
• It takes less time to train a machine-learning model
• Machine learning requires domain expertise to train whereas deep learning does not (feature selection is required in machine learning for example training a model to know what houses will be the best to buy, you will have to tell it the features that make houses more valuable. In the case of deep learning, the machine determines this by itself)
• With deep learning, you cannot tell how a model arrived at an output

Supervised, unsupervised and reinforcement learning

Supervised (show the machine what you want it to do - labelled data):

In this type of learning, the machine is shown the output that it is supposed to achieve.

• teaching a machine how to identify a dog, labelled images of dogs and images of things that are not dogs are shown to the machine until the machine can identify a dog and a non-dog image.
• teaching a machine to tell which clients will default on their loans by feeding it millions of data on users that default on their loans.
• teaching a machine to predict trends by giving it historical data

unsupervised (Letting the machine identify patterns on its own — unlabelled data):

This is used when you do not exactly know what the output will be.

• letting the machine determine commonalities (clustering) between customers to group customer types

Reinforcement learning (tell the machine a goal and let it figure out how to achieve the goal — give the machine rules).

• it knows what the output should be and what the rules are