Copyright: ThalesSource : Thales

Challenge

AI x Secure Update

ApplyAppliquer
This challenge is over, applications are closed.

Challenge Overview

Description du challenge

There must never have been a time when the use of AI-enabled systems were more prevalent than today to optimize operational performances. In this regard, the ability to update AI capabilities while ensuring data systems security is one of our priorities.Therefore, the objectives of the challenge are to:

  • Secure and accelerate the update of AI between several actors (customers, Thales, suppliers...)
  •  Update product's AI function by a new learning phase including additional data with the constraint to evaluate the performance on reference data and/or to evaluate the initial model on a new dataset
  • Ensure security and limit the sharing of industrial data and customer data considered sensitive
  • Design a software platform (MLops/MLSECOps) and tools capable of updating AI-machine learning models with limited or no data/model sharing.The platform should allow different update scenarios between parties with various levels of transparency on model performance and/or dataset representativity
Copyright: Thales
Copyright: Thales
Copyright: Sergey Nivens
Copyright: Sergey Nivens

Wanted

Expert in:

  • Machine learning engineering
  • Data science
  • Optimization

Experience in various ML tasks such as:

  • Automatic target detection and identification
  • Computer vision
  • Predictive maintenance
  • Natural language processing
  • Develop an AI-software platform

Projects you could be working on

Implementation of a software platform and tools able to update models and perform machine learning without sharing data. The POC will be realized jointly by first deciding a particular task performed by ML (such as object detection, identification such as satellite images, predictive maintenance, natural language processing). A dataset will be identified to be used as a support for the POC (Thales data or public data). Then, a reference will be made with a scenario of total sharing of information between the parties, especially of data. It will be required to define one or more scenarios of updating the model with limited sharing. Finally, the results will be compared between the reference scenario (total sharing) and the scenario(s) with limited sharing.