idw – Informationsdienst Wissenschaft
Nachrichten, Termine, Experten
Nachrichten, Termine, Experten
The Institute of Science and Technology Austria (ISTA) has launched its Responsible AI Program, a new initiative made possible through the Ꞓ5 million donation by Canadian entrepreneur Garrett Camp announced earlier this year. The program supports cutting-edge research that advances artificial intelligence as a reliable, human-centered technology. The first selected projects address key challenges in privacy, transparency, robustness, and safety—core pillars of responsible AI.
Made possible through a donation from Canadian entrepreneur and philanthropist Garrett Camp, co-founder of Uber, the program funds PhD students, postdoctoral researchers, scientific interns, and residents working on projects that promote fairness, transparency, privacy, and accountability in Artificial Intelligence (AI).
The first call for applications received a strong response from early-career scientists across disciplines. Following a competitive evaluation by the Responsible AI Committee, comprised of experts inside and outside ISTA, four postdoctoral researchers were now selected for the program’s inaugural cohort and are starting their projects.
Pioneering research for responsible innovation
The four newly awarded projects embody ISTA’s interdisciplinary approach to fundamental AI research and its commitment to developing technologies that serve society responsibly.
Joel Daniel Andersson, postdoc in the research group of Monika Henzinger, will work on the “Streaming Differential Privacy: Efficient and Explicit Mechanisms” project. From health apps to language models, today’s AI systems often rely on vast amounts of personal data, raising critical questions about how to protect individual privacy. Andersson’s research tackles this challenge by developing new ways for AI models to learn from sensitive data without exposing it. His project focuses on designing memory-efficient and mathematically sound methods for differential privacy—ensuring that private information remains secure while keeping models accurate and useful.
Filip Cano, postdoc in the research group of Thomas Henzinger, will work on the “Adaptive Enforcement of Safety and Fairness of AI Decision Makers at Runtime” project. Many modern AI systems operate in settings where their decisions can have significant consequences and must adapt to changing conditions. Cano’s research explores how such systems can be supervised and adjusted while they are running, ensuring their behavior remains safe and fair in real time. His goal is to design explainable, flexible interventions that correct potential issues without unnecessary disruption—making AI both more reliable and easier for humans to understand and trust.
Valentino Maiorca, postdoc in the research group of Francesco Locatello, will work on the “Input-Conditional Inference on Frozen Transformers” project. Large AI models, such as those behind chatbots or image generators, activate thousands of internal components every time they process an input—even when many of those parts are irrelevant. Maiorca’s research investigates how to identify which modules are actually needed for a given task or input, allowing others to be skipped. This approach can make AI systems more understandable, show when their predictions may be unreliable, and reduce their energy use by avoiding unnecessary computation.
Marco Pegoraro, postdoc in the research group of Alexander Bronstein, will work on the “Adversarially Robust Confidence Estimation for AI in Computational Biology” project. Biologists increasingly rely on AI models to predict the 3D structures of proteins. These specialized models — such as AlphaFold — therefore deliver crucial steps in developing new drugs and therapies. These systems also provide confidence scores indicating how trustworthy each prediction is, but those scores can sometimes be misleading. Pegoraro’s project examines why this happens and develops methods to make AI confidence measures more robust, helping researchers better decide which predictions to trust and ultimately improving the reliability of AI-driven discoveries.
A forward-looking program for long-term impact
“The Responsible AI Program reflects ISTA’s vision of advancing AI research that is both scientifically rigorous and socially meaningful,” says ISTA President Martin Hetzer. “Thanks to Garrett Camp’s generous support, we are empowering young scientists to explore groundbreaking questions about how AI can be designed, understood, and deployed responsibly.”
The program’s three annual calls will ensure continuous opportunities for ISTA early-career researchers to contribute to this mission. Each selected project exemplifies the Institute’s unique culture of collaboration across computer science, mathematics, and the natural sciences.
This initiative builds on the donation announced in January, which marked Camp’s first philanthropic gift to a European research institute. This underscores a shared vision for responsible innovation and global collaboration in AI. In this context, as AI systems increasingly shape scientific discovery and everyday life, ISTA’s Responsible AI Program provides a crucial contribution to ensure that these technologies develop in harmony with human values.
https://ista.ac.at/en/news/ista-receives-e5-million-donation-for-ai/ ISTA Receives €5 Million Donation for AI Research by Uber co-founder Garrett Camp
Postdocs Joel Daniel Andersson, Filip Cano, Valentino Maiorca, and Marco Pegoraro (from left to righ ...
Copyright: © ISTA
The Institute of Science and Technology Austria (ISTA) has launched its Responsible AI Program. It s ...
Copyright: © ISTA | Magic Lemur Productions
Merkmale dieser Pressemitteilung:
Journalisten
Informationstechnik
überregional
Forschungsprojekte
Englisch
Sie können Suchbegriffe mit und, oder und / oder nicht verknüpfen, z. B. Philo nicht logie.
Verknüpfungen können Sie mit Klammern voneinander trennen, z. B. (Philo nicht logie) oder (Psycho und logie).
Zusammenhängende Worte werden als Wortgruppe gesucht, wenn Sie sie in Anführungsstriche setzen, z. B. „Bundesrepublik Deutschland“.
Die Erweiterte Suche können Sie auch nutzen, ohne Suchbegriffe einzugeben. Sie orientiert sich dann an den Kriterien, die Sie ausgewählt haben (z. B. nach dem Land oder dem Sachgebiet).
Haben Sie in einer Kategorie kein Kriterium ausgewählt, wird die gesamte Kategorie durchsucht (z.B. alle Sachgebiete oder alle Länder).
