Gustavo Carneiro
Full Professor of AI and Machine Learning, University of Surrey
Gustavo Carneiro is a Professor of AI and Machine Learning at the University of Surrey, UK. Before that, from 2019 to 2022, he was a Professor at the School of Computer Science at the University of Adelaide, an ARC Future Fellow, and the Director of Medical Machine Learning at the Australian Institute of Machine Learning. He joined the University of Adelaide as a senior lecturer in 2011, has become an associate professor in 2015 and a professor in 2019. In 2014 and 2019, he joined the Technical University of Munich as a visiting professor and a Humboldt fellow. From 2008 to 2011 Dr. Carneiro was a Marie Curie IIF fellow and a visiting assistant professor at the Instituto Superior Tecnico (Lisbon, Portugal) within the Carnegie Mellon University-Portugal program (CMU-Portugal). From 2006 to 2008, Dr. Carneiro was a research scientist at Siemens Corporate Research in Princeton, USA. In 2005, he was a post-doctoral fellow at the University of British Columbia and at the University of California San Diego. Dr. Carneiro received his Ph.D. in computer science from the University of Toronto in 2004.
Areas of specialism
Lecture
Learning to Complement and to Defer to Multiple Users using Personalized Human-AI Collaborative Classifiers
Medical image classification tasks can be complex, with expert labellers sometimes unsure about the classes present in the images. This uncertainty leads to the issue of learning with noisy labels (LNL). The ill-posed nature of LNL requires the adoption of strong assumptions or the use of multiple noisy labels per training image. These methods can result in accurate models that perform well in isolation but may not be as reliable as the consensus classification by humans. An alternative approach leverages the synergies between human expertise and AI capabilities in human-AI collaborative classification (HAI-CC). However, the complex decision-making process involved in exploring these synergies makes HAI-CC a challenging task. HAI-CC includes three options: 1) AI autonomously classifies; 2) learning to complement, where AI collaborates with users; and 3) learning to defer, where AI defers to users. Despite their interconnected nature, these options have been studied in isolation rather than as components of a unified system. Furthermore, in the learning to complement option, the AI model often oversimplifies the problem by relying on a global human behaviour model, which overlooks human variability, leading to sub-optimal solutions. In this lecture, we will introduce concepts such as learning with noisy labels, multi-rater learning, and human-AI collaborative classification. We will then discuss the latest developments from our lab in HAI-CC, particularly a system that unifies the three options of HAI-CC and a learning to complement system that can personalise to individual human behaviour models.
