Seminar: Beyond the Tracks: On the Necessity of Sample-Efficient, Robust, and Adaptive AI

As deep learning approaches the limits of sustainable data scaling, this presentation advocates for a transition from systems best described by a train analogy characterized by structural rigidity and infrastructure reliance to a rover-inspired paradigm defined by sample efficiency, robustness , and resilience. We challenge the prevailing trend of increasing model complexity through large scale empirical evaluations across domains like image classification and time series forecasting. These studies demonstrate that rigorously tuned baselines frequently match or exceed the performance of complex architectures while remaining more computationally efficient. Furthermore, we show that alternative regularization strategies can effectively replicate the benefits of data scaling in regimes where samples are scarce. The necessity of adaptive systems is further illustrated through results obtained in clinical scenarios involving personalized diabetes management, where real-time adjustment is required to sync with evolving physiological dynamics.

We finally discuss future research avenues concerning the application of how this rover-based framework supports the development of digital twins by addressing core requirements such as sample efficiency, robustness, and continuous evolvability in shifting environments.

Dr Lorenzo Brigato is a Postdoctoral Researcher at the ARTORG Center for Biomedical Engineering Research at the University of Bern, where he focuses on the application of artificial intelligence to healthcare. He holds a PhD in Computer Science and an MSc in Artificial Intelligence and Robotics with honors from Sapienza University of Rome, as well as a BSc in Engineering Sciences from Tor Vergata University of Rome.

His research primarily addresses core challenges in modern machine learning: improving the data efficiency of learning models and enhancing their robustness and adaptability. Throughout his career, he has presented his work at international conferences in computer vision and robotics and has collaborated with researchers from institutions such as the University of Verona, Friedrich Schiller University Jena, and Delft University of Technology.

In addition to his research, he has taken on various academic leadership and mentorship roles, including the co-supervision of Master's theses and the delivery of specialized seminars and tutoring. His background also includes a research stay at the HES-SO University of Applied Sciences and Arts Western Switzerland, where he developed machine learning algorithms for motion recognition in robotic hand prostheses.