*** Addressing uncertainty in the safety assurance of machine-learning *** 💡 When deploying machine-learning-based systems in applications such as autonomous vehicles or medical devices, ensuring that errors in the learned models do not lead to unacceptable risk is of paramount importance. However, formulating credible safety assurance arguments for such systems is challenging and becomes rapidly more difficult as system complexity increases. The resulting lack of confidence in the assurance argument can be characterised by various sources of uncertainty. These include an incomplete understanding of the complex, evolving operating environment and often ambiguous nature of the task to be performed, the difficulty in collecting complete and representative data for training and test, and the inherent uncertainty within the machine learning models themselves. 📄 In this first paper of their latest strand of research into AI safety assurance, Simon Burton and Benjamin Herd present an uncertainty-aware assurance methodology for AI-based systems to increase the level of trust we have in our assurance arguments and thereby in the systems themselves. In particular the methodology addresses the following aspects of the assurance argument: 👉 How to identify factors that might undermine trust in quantitative evidence (e.g. ML metrics) and how might this evidence best be combined to reduce uncertainty. 👉 Understanding the extent to which invalid assumptions could potentially undermine the safety claims of the argument. 👉 How to combine quantitative arguments with qualitative evidence to build a robust assurance approach that is continually refined to reduce residual and emerging uncertainties after the function has been deployed into the operating environment. 🔗 Read the full paper: https://lnkd.in/d2w-BQDk 📚🤓 Want to learn more? Their second paper on this topic will be presented soon. Further work will be presented at this year's SafeComp (September 17-20, 2024, Florence). 🔗 Registration is now open ⏩ https://lnkd.in/dPUEMcKN #safetyassurance #safety #machinelearning #SafeIntelligence
Fraunhofer IKS
Forschungsdienstleistungen
Wir bringen intelligente Software-Technologien und Safety in Einklang. Für eine Welt, auf die wir uns verlassen können.
Info
Das Fraunhofer IKS arbeitet an innovativen Software-Ansätzen, die technischen Systemen die benötigte Intelligenz unter höchsten Qualitätsansprüchen verleihen. Durch ihre Kompetenz in den Bereichen Adaptive Systems, Dependable Systems und Quality of Service tragen die Forscherinnen und Forscher damit auch maßgeblich zum Gelingen sicherheitskritischer Anwendungen für Bereiche wie Industrie 4.0 oder Mobilität der Zukunft bei. Der Forschungsschwerpunkt des Instituts liegt auf resilienten kognitiven Systemen, die durch ihre verlässliche Funktionsfähigkeit Voraussetzung für digitale Zukunftsvisionen wie autonome Systeme und deren Zusammenspiel sind. Institutsleitung: apl. Prof. Dr. habil. Mario Trapp Impressum: https://www.iks.fraunhofer.de/de/impressum.html
- Website
-
http://www.iks.fraunhofer.de
Externer Link zu Fraunhofer IKS
- Branche
- Forschungsdienstleistungen
- Größe
- 51–200 Beschäftigte
- Hauptsitz
- München
- Art
- Nonprofit
- Gegründet
- 1999
- Spezialgebiete
- Embedded Systems, Sichere Vernetzte Systeme, Industrie 4.0, Kooperative Fahrerassistenz, I4.0, Industriekommunikation, Autonomes Fahren, Künstliche Intelligenz, Automotive, Adaptive Systems, Dependable Systems, Quality of Service, resiliente Intelligenz, Industry 4.0, Autonomous Driving, Artificial Intelligence, Machine Learning, Safe Intelligence und Kognitive Systeme
Orte
-
Primär
Hansastr. 32
München, 80686, DE
Beschäftigte von Fraunhofer IKS
-
Thomas Messerer
-
Stefano Gianolini
Sabbatical: Quantum Computing in Healthcare
-
Karsten Roscher
Safe Intelligence for Industrial Sensing Systems | Head of Department at Fraunhofer IKS
-
Reinhard Stolle
Director Business Unit Mobility at Fraunhofer Institute for Cognitive Systems
Updates
-
Today at the IKOM in Garching/Munich: Visit us at booth 61 - our HR team with the active support of our scientists will be happy to inform you about exciting career opportunities at Fraunhofer IKS. 🤩💪💯 Cool team and great chances for you - join us! 🚀 Kirsten Plesch | Erik Peter | Poulami Sinhamahapatra | Alexandre Sawczuk da Silva | Theresa Demmelhuber #SafeIntelligence #career #jobs #experience
-
-
AI in medicine? Sure, but how is that supposed to work? 🧐 Get exciting insights in our training "AI in Medicine" on July 3, 2024 at Fraunhofer IKS! 📅 👉 Where and how can AI be usefully deployed in hospitals, health IT applications and other healthcare areas as of today? 👉 How can the trustworthiness, safety and verification of AI models be adequately taken into account using state-of-the-art approaches, from the idea to the actual application? 👉 What are current limitations of AI in medicine, and how can these be addressed? 👉 How to deal with challenges such as "(too) little data"? 🔜 Register now until next Friday & be part of it! You can find more information here: https://lnkd.in/dgbJPamj #SafeIntelligence #healthcare #medicine #ai #training
-
📣 Exciting update on this year's Safetronic: the time has come, the committee selected the best from numerous, very interesting papers ... and we would now like to give you a first insight into the program of Safetronic 2024! 🤩 Find out in an interview with Frank Kirschke-Biller - member of the program committee - his view of the conference in November and read about the inspiring presentations he is looking forward to! 💯 🔗 Click here for the interview: https://lnkd.in/dqDCQNHD 🔗 Registration is now open 🎟️ Don't want to miss the conference on the holistic safety of road vehicles? Then register right here: https://lnkd.in/d4YKkPZA Mario Trapp | Håkan Sivencrona | Nicolas Becker | Simon Fürst | Stefan Kriso | Peter Lascych | Andreas Knapp | Frank Kirschke-Biller | Hans-Leo Ross | Thomas Frese #Safetronic24 #safety #ai #conference #SafeIntelligence
Interview Frank Kirschke-Biller – »The holistic view of product safety is very important to me« – Magazine of the Fraunhofer Institute for Cognitive Systems IKS
safe-intelligence.fraunhofer.de
-
During our two day EU AI Act training, we covered an overview of the EU AI Act and the IKS verification framework, providing our participants with a forward looking perspective on the implications of this regulation for their businesses. 🤓💪 We had a great group of enthusiastic and curious participants which led to very interesting discussions! 💯 Thanks to our guests and speakers! Jessica Kelly | Lena Heidemann | Anna Guderitz | Josef Jiru | Alireza Zamanian #eu #ai #training #SafeIntelligence
-
-
Dr. Reinhard Stolle has been appointed Deputy Director of the Fraunhofer IKS - we are particularly pleased about this news! 💯 With Reinhard, we have an excellent analytical mind and experienced manager at our side who knows the market like few others. 🚀 He has been at Fraunhofer IKS since the beginning of December and successfully heads the Business Unit Mobility. Congratulations Reinhard & good luck in your new position! 🤗 #SafeIntelligence
-
-
Vom Himalaya ins Alpenvorland ⛰️⛰️: wie Dr. Florian Geissler ans Fraunhofer IKS gekommen ist und was er hier macht, lesen Sie im Porträt auf dem Safe Intelligence Blog: »Gerade sehen alle mehr und mehr, was durch KI alles möglich ist!« 🧐 Hier geht´s zum Artikel - viel Spaß beim Lesen 📖: https://lnkd.in/d6VAJXNX #SafeIntelligence #ki #quantumcomputing #safety
Porträt Florian Geißler – »Gerade sehen alle mehr und mehr, was durch KI alles möglich ist!« – Magazin des Fraunhofer-Instituts für Kognitive Systeme IKS
safe-intelligence.fraunhofer.de
-
You want to learn something more about »AI in Healthcare«? 🩺🤓 🔜 Join our training on July 3, 2024 in Munich and get registered here: https://lnkd.in/dpENAwHf 🔗 Find some more information about the training here: https://lnkd.in/dgbJPamj #SafeIntelligence #healthcare #medicine #ai #training
-
Yesterday the call for papers for Safetronic 2024 ended and we would like to thank everyone for their submissions. 💯 We received a lot of very interesting papers! 📃📃 The program committee is now reviewing the papers - next step is to select the most suitable ones for the program 2024. 🤩 🔜 Stay tuned for a first look at the agenda in June! Mario Trapp | Håkan Sivencrona | Nicolas Becker | Simon Fürst | Stefan Kriso | Peter Lascych | Andreas Knapp | Frank Kirschke-Biller | Hans-Leo Ross | Thomas Frese #Safetronic24 #safety #ai #conference #SafeIntelligence
-
-
☑️ Was ist Quantencomputing und wie kann es in realen Anwendungen helfen? ☑️ Wie ist der Stand der Technik bei Quantencomputing und was kann ich für die nahe Zukunft erwarten? ☑️ Welche Anwendungsbereiche gibt es? Welche Vorteile ergeben sich daraus? ☑️ Wie beginne ich mit der Entwicklung quantengestützter Methoden für meine Anwendungsfälle? Auch Sie benötigen Antworten auf diese Fragen? ⏩ Dann sind Sie bei uns genau richtig: diese und mehr beantwortet Ihnen u.A. unser Wissenschaftler Benedikt Poggel am 7. Juni 2024 vor Ort am Fraunhofer IKS! 🚀 🧐 Nutzen Sie unser exklusives Training als Chance für sich und Ihr Unternehmen und melden Sie sich bis spätestens 22. Mai direkt hier an: https://lnkd.in/eJ5EBJjY #ai #quantencomputing #potenzial #training #SafeIntelligence
-