Fujitsu’s TDA Technologies

Fujitsu’s TDA Technologies Shaping features with you

Welcome to Fujitsu’s TDA Technologies!!
In this page, we introduce our new technology based on topological data analysis (TDA)
for analysing complex time series data.

What’s TDA

Topological Data Analysis (TDA) is a method characterizing geometric features inherent in a dataset.
Fujitsu have devised a novel method based on TDA for analysing complex time series data.

What’s TDA

  1. Input: Input time series data to be analysed. For complex time series data, it is difficult for us to determine whether noises or anomalies exist or not by just seeing the original signal.
  2. Embedding: Convert the input signal into a point cloud in the Euclidean space via time-delay embedding.
    The trajectory drawn by the embedding (pseudo-attractors) characterizes the generation process of the input signal as a geometrical shape.
  3. Feature Extraction: Apply TDA to the point cloud and extract features based on its geometrical shape.
  4. Training and Analysis: Use the features output by the TDA for classification, anomaly detection, or other machine learning tasks.

Advantages of Our TDA Technologies

Our TDA technology is robust to noise included in signals because it extracts features based on the data generation process.
Conventionally, we have to manually preprocess the time series data monitored from sensor devices (such as vibration sensor or electroencephalograph) to remove noise.
However, when we apply our TDA to time series data, it can be analyzed without such preprocessing.

With our TDA, Fujitsu is developing technology for detecting defection in road bridges using vibration sensors, and technology for early detection of signs of illness using electroencephalography.

Case Study

Our TDA technologies have a capability of dealing with a wide range of fields.
The following are some use cases of our TDA.

  • CASE 1

    CASE1Motion detection with accelerating data

    Classify what kind of motion a person is performing by using accelerometers attached to the arms, legs, and hip.

  • CASE 2

    CASE2Delirium detection using EEG

    Detecting whether delirium occurs or not with EEG data obtained by an electroencephalograph.

  • CASE 3

    CASE3Internal defect detection of a bridge

    Detect internal defect of the bridge based on the vibration data the sensors attached on the surface of a bridge.

About Us

Currently, Artificial Intelligence Laboratory of Fujitsu has formed a research group on TDA and is carrying out research and development activities.

Our TDA team mainly consists of two teams: "Fundamental Research Team" and "Applied Research Team".
The former is mainly responsible for the development of basic and advanced technologies for TDA.

On the other hand, the latter is responsible for promoting the commercialisation of TDA. Our group is also collaborating with INRIA (Institut National de Recherche en Informatique et d'Automatique), a French national research institute and some of the results of this collaboration are implemented in the TDA OSS library, Gudhi.()




Time Series Shaper

In order to try our TDA technologies, we have released a service that allows you to easily use time series analysis with TDA.

This service offers two modes: AUTO and EXPERT, which allow users to upload time series data and learn a model for extracting features, and TEST, which allows users to extract features from new time series data using the created model.

For detailed instructions, please click and see the following page.