Where are we leaping from?

The goal of AI is to make machines mimic the way humans think.

• Understanding the concept of AGI (Artificial General Intelligence) and human cognition
• Framing AI with Computer Vision and Reinforced Learning
• Enabling AGI and shared learning by combining technologies

Abzu technology is based around Quantum Lattice (QLattice) and breaks new fields:

• Transparent model – Explain the ‘Why’ to enable insights
• Shared learning – Shared model building – small local data sets / ‘save’ model learnings
• Small Local Datasets – Ease of use with less data preparation and easy model deployment

Quantum Lattice accumulate learning of a structure through the perception of reality. As it is fed more data it gets smarter at solving problems that ascertain to the data in a generalised and cognitive way.

Triggering Human Cognition: A simple question will do it!

How many red cars are there in Spain? As soon as you read or hear the question your brain has already started the cognition process – personal experience, historic facts, visits to Spain, etc. Although you have no real data, you will probably have a number in mind.
Fact: There are around 600,000 red cars in Spain.

As your expertise grows your ability to solve a problem through cognition increases. A computer thinks very differently to human cognition.
ConvNets – Convolutional Neural Networks

• Computer vision, object detection and object localisation.
• 14,000,000 images labelled by humans

However, ConvNets are easily fooled.

Reinforcement learning – the approach is through simulation and repartition. It’s a simple computer game that it replays 100 episodes – low score, 100,000 episodes – very good score, millions of episodes – and it beats any human player. If any parameters change then machine goes back to the beginning.
This means there’s no real cognition, so every possible change must be simulated.

QLattice can predict demand on a construction site.

Casper Wilstrup

How can we enable AGI and shared learnings?

• Locate datasets – small and local datasets
• Data is fitted to the surface of a QLattice to enable shared learning
• Run on a supercomputer – automatically locate and present Transparent Model – easy to explain ‘why’ and deploy insights
• Model learnings will be stored in the QLattice

QLattice helps understand what drives demand, therefore predicting events is achieved with high accuracy. This is very useful for predicting demand for machines, etc. in the construction industry.

Watch the full keynote here: