Mathematics and Technology
In 2016, Via Science released our microservices architecture, Focus™. Microservices architecture is a growing trend in the industry and our implementation improves the handling of varying types and frequency of data as well as differences in operational deployment.
Designed around the Kappa Architecture, Focus™ uses the same code to train, test and deploy models, so it processes streaming and batch data equally well. The architecture isolates one micro piece of robust software without affecting other pieces. This allows us to create smaller, independent modules and therefore, customize our approach for each client.
Focus™ uses Apache Avro – a data description schema – to ensure easy communication and integration with other data sources regardless of the operating system. The architecture publishes outputs in standardized formats like JSON and REST API to make the final algorithms lightweight and deployable on a wide range of infrastructures.
At the core of Focus™ is Via Science’s patented machine learning engine, REFS™ (Reverse Engineering and Forward Simulation). The engine automates causal inference, which creates more accurate predictive algorithms and excels at addressing diagnostic and simulation problems. Built on Bayesian networks – a new branch of mathematics – this engine is particularly well-suited to handle missing and sparse data: a major problem in real world applications, such as with power grids or in cybersecurity.
What makes Via Science’s approach to predictive analytics different?
At Via Science, we do more than just make predictions. Using Focus™, we detect anomalies and infer cause and effect relationships to help clients choose the best action to minimize risks. We automate our analysis using Big Math, so it’s faster, repeatable and you don’t need an army of specialists.
What does the term Big Math mean?
Big Math is Bayesian mathematics (the Math) optimized on a massively parallel supercomputer (the Big). We use Big Math to analyze big, diverse and sparse data to predict what might happen in the future and how you can accelerate or attenuate potential outcomes.
What are the benefits of using Bayesian mathematics?
We use two types of Bayesian mathematics: statistics and networks. Recent studies have shown that Bayesian statistics is a proven way to solve for the problem of overfitting. Bayesian networks excels when you want to determine cause and effect relationships between data (rather than just correlations), and understand what you can do differently to achieve new outcomes.
If Bayesian mathematics is so good, why doesn’t everyone use it?
Other companies are starting to use Bayesian mathematics. But, there are two key reasons this approach is not more common: the limited number of Bayesian experts, and the big computing power required. We developed our platform so you don’t have to be a Bayesian expert or have access to big computer to use this approach.