Return
Thales

Thales

Anonymize defense data to train a machine learning model

Challenges

  • Anonymize sensor data for train a model of prediction
  • Desensitize data in a Defence context
  • Test the level reliability of data (multi-variate time series) on existing algorithms by comparing the level of utility

Maintaining statistical quality and utility

- Comparison of horizontal speed curves (VH) between original data and avatars - each curve represents a flight

‍

-Comparison of altitude curves (Alt) between original data and avatars - each curve represents a flight

Here we find the differences in correlation between pairs of variables - the values are close to zero so we keep the correlation well (for example between altitude and horizontal speed)

ROI

  • Optimizing predictive models : Precise reconstruction of time traces to improve the performance of data models.
  • Simplified and secure sharing : Expanded access to anonymized data, allowing faster exploitation and in compliance with regulations.
  • Increased visibility: Key results published in a white paper dedicated to AI and Defense, valuing the project and its strategic impacts. [Link coming soon]
“This collaboration represents a major challenge for the processing of strategic and confidential data in Defense. Together, we are exploring the application of their avatar anonymization method to Defense data, in particular for training Machine Learning algorithms. (...) We plan to share the results in a future white paper on AI and Defense.” - Marine Martinez, Program Lead CyberStationF, Thales
Download the use case
Our customers
All our customers
Wefight
Wefight
Analyses

Wefight, an innovative Healthtech company, is developing Vik. Vik is an artificial intelligence that supports patients with chronic diseases throughout their care journey. Through these various services, Wefight is required to collect and process a significant amount of personal and sensitive health data.

See the customer sheet
Sherbrooke University
Sherbrooke University
Open Data

The MeDomics research laboratory at the University of Sherbrooke, led by Professor Martin Vallières, has focused on the creation of predictive models in health informatics since its foundation in 2020.

See the customer sheet
Top Turnover
Top Turnover
IA gen

T.O.P, The Augmented Manager. With Top Turmover, anticipate resignations, identify future causes of departure, and visualize corrective actions over time to reduce your turnover rate.

See the customer sheet