SafeML based reliability assessment
Coordinated by
IESE
In earlier work, a statistical distance-based measure (SafeML) is proposed for machine learning components. In BIECO project, we propose extension of it with the use of Statistical Distance Dissimilarity across time series to obtain SDD based reliability and robustness estimate (StadRE and StadRO).
Description
Describe the innovation content of the result:
Obtain the statistical distance dissimilarity based reliability assessment which takes into account the distance between the data distributions and the performance.
Who will be the customer?
The customers are researchers, industry that want to assess the reliability of their ML component.
What benefit will it bring to the customers?
Runtime prediction of the ML component reliability.
When is the expected date of achievement in the project (Mth/yr)?
Methodology concept in 09/2022, SafeML based reliability 08/2024
When is the time to market (Mth/yr)?
May need some more maturity after the end of Project
What are the costs to be incurred after the project and before exploitation?
SafeML based reliability assessment will be ready for certain use cases without further investment after finishing BIECO but further generalization and research based on it will need.
What is the approximate price range of this result/price of licences?
Still in discussion-
What are the market size in Millions € for this result and relevant trend?
ML market size was estimated to be ~4 Billion in 2021. Expected to grow upto 120 billion by 2030.
How will this result rank against competing products in terms of price/performance?
The addresses the use of ML in safety/financial critical system where a runtime reliability estimate is needed..
Who are the competitors for this result?
National and international public and proprietary evaluators that implement a runtime simulation-based approach for assuring the safety of self-adaptive systems..
How fast and in what ways will the competition respond to this result?
Creation of a similar solution will take at least three years..
Who are the partners involved in the result?
The concept was evaluated on the use case by 7Bulls
Who are the industrial partners interested in the result (partners, sponsors, etc.)?
7Bulls is interested in the results.
Have you protected or will you protect this result? How? When?
Research publications are provided along the way of the concept development and its validation.
Other results
Security Testing Tool
GdpR-based cOmbinatOrial Testing (GROOT) is a general combinatorial strategy for testing systems managing GDPR’s concepts (e.g., Data Subject, Personal Data or Controller).
Domain Specific Language
The domain specific language enables specification of digital twin behaviour in a manner that can enable a predicted evaluation of its trustworthy behaviour in a simulated environment
Fail-operation clock synchronization methodology
Synchronization loss can occur due to many reasons, either because of a device or link failure or due to a targeted attack on the reference node, which supplies the corrected time to the network’s nodes.
Time sensitive network simulation
Simulate the real time communication for the distributed based on the Time sensitive network simulation. Additionally, simulate the fail-operation clock synchronization methodology. This simulation will validate the communication stack.
Remote updating- upgrading of vehicle firmware
Securing remote and in-vehicle communications against cyber-attacks, possibly performed with quantum computers in the near future, is a major goal in the automotive sector.
Extension of the approach to the communications within the whole supply chain of the Microfactory
Once the secure communication with a vehicle has been demonstrated for remote FW update, it is straightforward to extend it to a general Service-Over-the-Air architecture and even further.
