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.