Introduction to Pattern Computer

Introduction to Pattern Computer

Pattern Computer is focused on discovering patterns in complex, high dimensional datasets. We know from experience that many entities have endless petabytes of big data where hidden patterns can be found that capture significant strategic and monetary value for its custodians. Existing machine learning methods are not particularly useful when applied to discovering new patterns within datasets. Thus, a novel approach must be taken to make these previously-unseen patterns emerge – so far, the computational horsepower and/or mathematical complexity necessary to find the patterns within diverse datasets have been unsurmountable problems. Pattern Computer has solved these problems. Our supervised machine learning approaches allow us to discover and identify the key factors in a dataset most responsible for specific outcomes. We can then use these factors to build mathematical representations of the data to model the relationship. We can model specific outcomes and make predictions on which factors to change/monitor to optimize it. Our unsupervised machine learning approaches allow us to identify clusters of similarities within the data. We can identify variables that are redundant, have similar function or representations and can be investigated for potential substitution, using a less expensive or more readily available component, element, or drug for example to accomplish the same goal. We can also identify variables that are complimentary or antagonistic, where we know that optimal function of a system requires deep coordination between these variables. While our initial focus of development has been discovering the genomic patterns behind diseases, these tools are dataset agnostic and have been used to identify issues in energy production emissions, manufacturing defect analysis, flight departure delays, as well as breast cancer.

Contrary to most companies using machine learning (of which most are using neural networks), Pattern Computer has developed our own proprietary algorithms to identify the critical patterns in high-dimensional datasets and are not computationally limited by the O(n2) scaling issues of neural networks (where n is the number of factors). In addition to pattern discoveries and building accurate mathematical models of the relationships between the factors, Pattern Computer also presents the relationships between these key variables in up to 8 dimensions via our Dimensional Navigator presenting the data in virtual reality. Pattern Computer is a dedicated group of approximately 20 accomplished experts in the areas of advanced mathematics, microbiology, computational algorithms, control systems, applied physics, hardware and software engineering, bioinformatics, data science, computer vision, visualization, and security.

Read the paper here.