AML is a patent-pending innovative Machine Learning (ML) algorithm. It was initially designed for online and real-time applications, suitable for modeling dynamic processes with uncertainties.

​

 

 

 

Big data contain a large number of data cells. Detecting the relationships between these data cells is the objective of any ML. Inherently, all data contain elements of uncertainties, and when data abstractions are performed the elements of uncertainty increase. A high degree of uncertainly may lead to a defective ML, where it accurately models the data – not the process.  

It deals with two types of data: (1) sequential data, in which every new data sample depends on past samples and affects subsequent samples, and (2) non-sequential data, in which each data sample is independent and is not affected by past sample nor does it impact on future samples.

AML is based on the fact that any historical dataset, in addition to explicit numerical values, contains valuable information in its statistics that can be used to extract relevant knowledge for creating a more relevant predictive model.

It was found that by using the proper statistical parameters of the dataset, a machine learning algorithm can be simpler and with fewer iterations and produce a predictive model that can cover a wide range of non-linear eventualities.

Unlike Neural Networks, AML does not contain any hidden element, no hidden layers and no hidden mode. 

This allows us to provide the user with data on every aspect of the learning algorithm, including how weight changed through and the error was reduced during the training sessions.