There are three key approaches to extracting Knowledge from Data:

Deep Learning – Using NEURAL NETWORK and MACHINE LEARNING algorithms to capture knowledge hidden within BIG DATA.

Forecasting – Using past data to make the best possible estimate of a future outcome. ARIMA based techniques are excellent for this.

Simulation – Using what you already know about the way elements of a problem interact to make a probability based estimation of future outcomes. Monte Carlo Simulation is excellent for this.

Quite often the optimum solution will involve hybrid models using all three technologies.

AI and Deep Learning is the current name for a collection of computational techniques that have emerged from Data Science labs since the early 1980’s.

The first generation of tools, such as OPS5, embodied the concept of Non-Prodedural Programming and Expert Systems. These tools are excellent for problems involving configuration, but of limited applicability to wider domains.

The second generation of tools emerged in the 1990’s. These tools were more useful for general problem solving and form the basis of many of today’s advanced Data Science workbench tools such as KNIME.

The second generation of tools were of two “flavours”, Rule Induction tools such as C4.5 and Artificial Neural Networks. These tools, though conceptually brilliant, were often of limited in practical use until today’s powerful compute platforms became available.

Deep Learning represents the current generation of Artificial Neural Network tools as developed and used by such organisations as Amazon, Google and Microsoft to power their analytical platforms at scale.

Though these tools are extremely effective they can require very special hardware platforms to run on and it is often only feasible to develop these models on workstation class hardware before  transferring them to a production environment such as AWS.

Even at AWS scale some problem domains are beyond today’s compute platforms but may become practical as Quantum Computers become available.

Monte Carlo Methods – a broad class of computational algorithms that rely on repeated random sampling to obtain numerical results.

These models are fitted to cross-impacting multi-dimensional data to better understand the data and to predict future outcomes (states).

Examples of use would be to predict:

Optimization of PPC Budget

Optimum Pricing to Maximize Revenue

Project Management Cost Estimation vs Budget

Portfolio Optimization

For those with a technical interest in the Monte Carlo Simulation technique go to the following links:

MIT Monte Carlo Simulation Video Course Module

Monte Carlo Simulation Theoretical Definition