Artificial Intelligence

This is the broadest of all the domains relevant to Machine Learning and is a superset of Machine Learning. It covers any computer system that can perform tasks usually set aside for a person.

The following are all example types of AI:
Machine Learning
Rules Engines (if they can be run automatically)
Robotic Process Automation

AI consists of a number of areas of Information Technology that is providing improvements across a number of areas for people. This includes helping people get their emails from their phones, having an online shopping enquiry go to the right product manager and smart homes where speaking into your phone turns on lights in a house.

TIP integrates with a number of other computer systems that are AI enabled. TIP itself is an AI platform, however, it is a specialised AI platform that leverages the most complex form of AI, Deep Neural Networks. This enables it to solve highly complex problems for our customers.

Machine Learning is a subset of AI. It uses training algorithms to solve problems. It achieves this by identifying patterns in the data instead of following programmatic instructions.

There are many types of Machine Learning, including:

1. Supervised learning via traditional statistical modelling (such as line of best fit algorithms)
2. Unsupervised learning
3. Reinforcement learning
4. Deep Learning

Non-Deep learning ML has been delivering significant results for an extended period of time. This includes determining if a person should be approved a home loan. The appeal of these ML methods has been their ability to be explained and while they have some computational overhead, it has not been as significant as Deep Learning.

Over the last decade, Deep Learning has become the preferred method of ML as it delivers more accurate ML results and computational power of computers in the last decade have advanced to the point that the additional overhead is no longer prohibitive.

TIP leverages ML. However, in order to deliver the best results for our customers, traditional ML methods do not often deliver the desired results. Instead, TIP uses world leading Deep Learning methods, rather than traditional statistical ML methods.

Deep learning is a subset of the ML domain. It is a collection of methods that leverage deep neural networks (neural networks) to solve complex problems. These neural networks usually deliver the most accurate results of any form of Machine Learning, such as telling apart various vehicle types, or translating a person’s voice into text. They can also detect anomalies, such as detecting faults in a jet engine or that a powerline is ageing and needs replacing

Neural networks can also be combined with another type of advanced machine learning, reinforcement learning, to learn how to win games such as the game of Go.

Deep Learning has infinite types as each type is effectively a different neural network architecture. However, there are several common types (with state of the art version of each type being discovered all the time). This includes:
• Feed forward neural networks (FFNN)
• Convolutions neural networks (CNN)
• Recurrent neural networks (RNN)

In practice, Deep Learning provides the best performing ML results, especially where the data and/or the objective of the task is complex or difficult to achieve using human or other ML methods. This comes at the cost of computational power required to make a decision. Recent advances in hardware and communications have significantly reduced this issue and as a result, Deep Learning is the ML domain of choice for organisations willing to achieve premium outcomes.

TIP leverages Deep Learning in its products. Not only this, but it leverages the world’s best Deep Learning methods and as such, provides the best results for any problem our customers face. Our unique ability to explain these results in real time, coupled with the ability to achieve these results without requiring significant data science capability, means our customers have the best of both worlds – the best results, without the complexity.