Notice: This Wiki is now read only and edits are no longer possible. Please see: https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/wikis/Wiki-shutdown-plan for the plan.
Difference between revisions of "OpenGENESIS WG"
(Added picture to the scope; updated structure image) |
|||
Line 11: | Line 11: | ||
openGENESIS drives and supports the investigation of an understandable, verifiable and certifiable AI. This includes the adoption of functional safety approaches to symbolic and subsymbolic AI, as well as the development of new methods and metrics for the verification of neuronal networks. Approaches regarding the assessment of performance, robustness and comprehensibility will be compiled. In addition, an open data set is planned to be shared to enable research based on realistic cases and safety relevant challenges. Finally, social concerns should also be considered relating to the certification of learned algorithms. | openGENESIS drives and supports the investigation of an understandable, verifiable and certifiable AI. This includes the adoption of functional safety approaches to symbolic and subsymbolic AI, as well as the development of new methods and metrics for the verification of neuronal networks. Approaches regarding the assessment of performance, robustness and comprehensibility will be compiled. In addition, an open data set is planned to be shared to enable research based on realistic cases and safety relevant challenges. Finally, social concerns should also be considered relating to the certification of learned algorithms. | ||
− | [[File:openGENESIS_scope_overview.png]] | + | [[File:openGENESIS_scope_overview.png|720px]] |
+ | |||
+ | == Examples for potential openGENESIS spotlight projects (GSLP) == | ||
+ | The participants of a spotlight project define their own development or research object. In order to give a better idea of possible goals within openGENESIS, the following examples of presumably relevant considerations are listed: | ||
+ | * Define data set interest and how they should be made available | ||
+ | * Establishment of an open and public available training and validation data set | ||
+ | * Methods for goal-oriented generation of synthetic data of behavior of vulnerable road users | ||
+ | * Identification of the necessary properties of synthetic data for an adequate real data representation | ||
+ | * Advances in understandable and explainable AI | ||
+ | * Find and define metrics for quantitative statements about performance and robustness | ||
+ | * Discuss possible adjustments of safety standards to fit for machine learning | ||
+ | * Platform based AI abstraction for better algorithm verification | ||
== Charter and Participation Agreement == | == Charter and Participation Agreement == | ||
Line 41: | Line 52: | ||
Matthis Eicher, TÜV Süd | Matthis Eicher, TÜV Süd | ||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
− | |||
== Past Events == | == Past Events == | ||
Line 62: | Line 62: | ||
More information: | More information: | ||
− | * [ Foundation Meeting Minutes] | + | * [https://wiki.eclipse.org/images/d/d8/20190722_openGENESIS_foundation_meeting_minutes.pdf Foundation Meeting Minutes] |
=== First meeting of the openGENESIS Working Group, TÜV Süd, Munich === | === First meeting of the openGENESIS Working Group, TÜV Süd, Munich === | ||
Line 72: | Line 72: | ||
This event will bring all interested parties together for information about the Eclipse Working Group idea. | This event will bring all interested parties together for information about the Eclipse Working Group idea. | ||
We will also discuss technical aspects which seem interesting for closer examination and start an open exchange. | We will also discuss technical aspects which seem interesting for closer examination and start an open exchange. | ||
− | |||
− | |||
Revision as of 08:55, 22 July 2019
Contents
Mission
openGENESIS is a collaborative platform with the mission to provide knowledge, methods and tools for the assessment of artificial intelligence (AI) that is used within autonomous driving applications. Before deployment onto public roads, learning algorithms must be proven safe and roadworthy. However, our current understanding of AI’s complex functionality is limited, especially in relation to machine learning algorithms. openGENESIS will provide both public and regulatory authorities with approaches to help them deal with the challenges of AI approval and certification. Providing a “TÜV for AI”.
Added value
The openGENESIS collaboration platform leverages knowledge among its members, enabling them to cooperate efficiently and share research results in an open access domain. This will establish a strong global exchange between industry, research and regulators to develop common criteria for the quality of AI.
Scope
openGENESIS drives and supports the investigation of an understandable, verifiable and certifiable AI. This includes the adoption of functional safety approaches to symbolic and subsymbolic AI, as well as the development of new methods and metrics for the verification of neuronal networks. Approaches regarding the assessment of performance, robustness and comprehensibility will be compiled. In addition, an open data set is planned to be shared to enable research based on realistic cases and safety relevant challenges. Finally, social concerns should also be considered relating to the certification of learned algorithms.
Examples for potential openGENESIS spotlight projects (GSLP)
The participants of a spotlight project define their own development or research object. In order to give a better idea of possible goals within openGENESIS, the following examples of presumably relevant considerations are listed:
- Define data set interest and how they should be made available
- Establishment of an open and public available training and validation data set
- Methods for goal-oriented generation of synthetic data of behavior of vulnerable road users
- Identification of the necessary properties of synthetic data for an adequate real data representation
- Advances in understandable and explainable AI
- Find and define metrics for quantitative statements about performance and robustness
- Discuss possible adjustments of safety standards to fit for machine learning
- Platform based AI abstraction for better algorithm verification
Charter and Participation Agreement
The openGENESIS Charter is drafted and available for review .
The openGENESIS Participation Agreement is also available.
Please contact ralph dot mueller at eclipse minus foundation dot org for more information.
Structure
openGENESIS provides a convenient and legal framework with lean and efficient structures. It is ensured by the Eclipse Foundation, which hosts the working group of openGENESIS. The collaborative platform also provides a flexible setting for openGENESIS spotlight projects (GSLP) to solve specific challenges in the community.
These projects can be initiated by one or more participant, taking a new aspect or improving approaches within the scope of openGENESIS. Partners retain their intellectual property, but are required to share their results and methodologies within the community. If the spotlight partners agree, spotlight projects can provide data to other projects or to the public. A steering committee manages the openGENESIS platform, defining and maintaining the scope and road map. It also decides on the acceptance and integration of new spotlight projects, and acts as a first contact.
Members
- DFKI (German Research Center for Artificial Intelligence) [DFKI]
- iMotion [iMotion]
- Incenda AI GmbH [Incenda AI]
- TÜV SÜD [TÜV SÜD]
Steering Committee
Steering Committee Chair
Matthis Eicher, TÜV Süd
Past Events
Foundation meeting of the openGENESIS Working Group, online
Date: Thu, July 18, 2019, 09:00 AM - 11:00 AM CEST
This meeting was the official foundation of the openGENSIS working group. Various bodies were set up and elected, a working mode for the steering committee was set up and requirements for the approval process of spotlight projects were discussed.
More information:
First meeting of the openGENESIS Working Group, TÜV Süd, Munich
Date: Tue, April 30, 2019, 9:00 AM – 4:30 PM CEST
From the invitation letter:
This event will bring all interested parties together for information about the Eclipse Working Group idea. We will also discuss technical aspects which seem interesting for closer examination and start an open exchange.
More information:
Presentations from the event:
- The GAP Between AI Development and Automotive Safety Criteria, C. Müller (DFKI)
- Eclipse Working Groups, R. Mueller (EF)
- Eclipse SUMO, R. Hilbrich (DLR)
- Proposal Spotlight Projects, Lucy Song (Imotion)
Participation
To participate in this community a minimum membership in the Eclipse Foundation as a Solution Member is necessary. Within openGENESIS two classes of membership are available:
Development Members
Development Membership is usually without an additional fee, but members are required to execute the openGENESIS Participation Agreement. They are also able to initiate or participate in spotlight projects under the umbrella of openGENESIS. To start a spotlight project, a project proposal must be submitted to the steering committee. As a minimum, this should contain:
- (Tentative) title of spotlight project
- Your contact information
- Participating entities
- Objective of the spotlight proposal
- Description of the approach to reach the objective
- Definition of planned data privacy
Driver Members
A Driver Member gives an annual fee and must execute the openGENESIS Participation Agreement. They can also initiate and participate in spotlight projects, and are additionally part of the steering committee.
Interested parties
- Mathworks [Mathworks]
- Phantasma Labs [Phantasma Labs]
- Bosch [Bosch]
- Eteration [Eteration]
- Intel Labs [Intel Labs]
Interfaces to other initiatives
Lernende Systeme – Plattform für Künstliche Intelligenz
Link: [https://www.plattform-lernende-systeme.de/home.htm]
Project type: German national funded project - by BMBF
Coordinator: acatech - Deutsche Akademie der Technikwissenschaften e.V.
Scope: Scope is to leverage knowledge and to point out perspectives in the area of artificial intelligence to position Germany internationally as a technology leader for learning systems. This should be reached by combination of the expertise from science, business and society.
Interface to openGENESIS:
Safety aspects – e.g. in the following groups
- AG 3 (IT Security, Privacy, Law and Ethic)
- AG 5 (Mobility and intelligent Infrastructure)
KI-Plattform - Konzipierung einer KI-Datenplattform zum Entwickeln und Testen autonomer Fahrzeuge
Link: [https://www.elektronikforschung.de/projekte/ki-plattform]
Project type: German national funded project - by BMBF
Coordinator: Volkswagen AG
Scope:
KI-Plattform will be used to design a common database for learning and test data, which will be available to industrial companies, research institutes and testing organisations as well. The project will design uniform and flexible data structures that can be used for current and future sensors. Beside real data, synthetic generated or modified data will also be considered. In addition, legal framework conditions and data protection aspects are examined and suitable operator concepts are developed.
Interface to openGENESIS:
AI Training and Testing Database is one of the major aspects of openGENESIS. Therefore, a close cooperation and coordination should be established and maintained.
VDA Leitinitiative - KI Absicherung
Link: tbd
Project type: German national funded
Coordinator: tbd.
Scope: tbd.
Interface to openGENESIS: Yes - tbd.