PerfectPattern
This website requires Javascript, to work properly.

PYTHIA

Automated Data Science

PYTHIA is a platform product for pattern recognition, time series prediction and anomaly detection on real-time data streams. Combining methods of deep learning, stochastic calculus, infinite dimensional geometry and quantum field theory, it autonomously finds even the most hidden patterns.

PYTHIA is a product for unsupervised and automated regression analysis, classification and time series prediction. It finds even the most complex patterns and relationships within unstructured and asynchronous data. It learns how to predict and control any sought-after quantity. There is no need for parameter tuning!

How PYTHIA works

PYTHIA can extract ordered data from many different streams of unordered factory data. This in turn can be used as a basis for factory control or to predict the probability distribution of a machine failure in the future.

Using PYTHIA will always be covered by the following four steps:

  • Stream unstructured and asynchronous data into PYTHIA
  • Tell it what you want to know, how it can influence the system and what it should accomplish.
  • It will autonomously learn how to answer your questions and how to control the system to achieve your goals.
  • The System autonomously learns how to improve quality and prevent failures.

Methodology

The following table compares the methodology of typical data analysis (others) step by step with PYTHIA.

  PYTHIA   Others
1 Unsupervised data synchronization 1 Data synchronization by data scientists
2 Unsupervised Anomaly Detection for Data Cleaning 2 Data Cleaning by data scientist
3 Unsupervised extraction of relationships including response times 3 Extraction of relationships including response times by data scientist and domain expert
4 Automated creation of model of time series 4 Root Cause Analysis and creation of model of time series by data scientist
5 Prediction of target quantity based on delay SDEs 5 Prediction of target quantity by data scientist
6 Learning of impact on expectation of target quantity 6 Creation of model to affect root causes by data scientist

You have any data set and you want to extract information from it? If the information is in there, PYTHIA will find it.


A new way of pattern recognition

The following tables compares pattern recognition of PYTHIA with artificial neuronal networks (ANN).

PYTHIA ANN
Amount of data needed Small Huge
Training global local
Calculation speed fast fast
Overfitting no tends to
Tweaking no a lot
Transparency full not clear why it works

Highlights

  • No need for parameter tuning or manual data cleaning.
  • If the information is in the data Pythia will find it (provable).
  • Searches in the space of ALL possible models. Even not restricted to ANNs.
  • Based on quantum field theory, stochastic processes and infinite dimensional geometry.

In conjunction with CORTEX there is thus the possibility that machines learn to independently counteract a failure.

© 2018 PerfectPattern GmbH