Keynotes

Mihaela van der Schaar is the John Humphrey Plummer Professor of Machine Learning, Artificial Intelligence and Medicine at the University of Cambridge and a Fellow at The Alan Turing Institute in London. In addition to leading the van der Schaar Lab, Mihaela is founder and director of the Cambridge Centre for AI in Medicine (CCAIM). Mihaela was elected IEEE Fellow in 2009. She has received numerous awards, including the Oon Prize on Preventative Medicine from the University of Cambridge (2018), a National Science Foundation CAREER Award (2004), 3 IBM Faculty Awards, the IBM Exploratory Stream Analytics Innovation Award, the Philips Make a Difference Award and several best paper awards, including the IEEE Darlington Award.

Mihaela is personally credited as inventor on 35 USA patents (the majority of which are listed here), many of which are still frequently cited and adopted in standards. She has made over 45 contributions to international standards for which she received 3 ISO Awards. In 2019, a Nesta report determined that Mihaela was the most-cited female AI researcher in the U.K.

AI for Science: Discovering diverse classes of equations in medicine and beyond
Abstract
Artificial Intelligence (AI) offers the promise of revolutionizing the way scientific discoveries are made and significantly accelerating their pace. This is important for numerous fields of study, including medicine. In this talk, I will present our research on AI for science over the past few years. I will start by briefly showing how we can discover closed-form prediction functions from cross-sectional data using symbolic metamodels. Then, I will introduce a new method, called D-CODE, which discovers closed-form ordinary differential equations (ODEs) from observed trajectories (longitudinal data).This method can only describe observable variables, yet many important variables in medical settings are often not observable. Hence, I will subsequently present the latent hybridisation model (LHM) that integrates a system of ODEs with machine-learned neural ODEs to fully describe the dynamics of the complex systems. However, ODEs are fundamentally inadequate to model systems with long-range dependencies or discontinuities. To solve these challenges, I will then present Neural Laplace, with which we can learn diverse classes of differential equations in the Laplace domain. I will conclude by presenting next research frontiers, including recent work on discovering partial differential questions from data (D-CIPHER).  While these works are applicable in numerous scientific domains, in this talk I will illustrate the various works with examples from medicine, ranging from understanding cancer evolution to treating Covid-19.
This work is joint work with Zhaozhi Qian, Krzysztof Kacprzyk and Sam Holt.

Polina Golland is a Sunlin (1966) and Priscilla Chou professor of Electrical Engineering and Computer Science at MIT and a principal investigator in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). Her primary research interest is in developing novel techniques for medical image analysis and understanding. With her students, Polina has demonstrated novel approaches to image segmentation, shape analysis, functional image analysis and population studies. She has served as an associate editor of the IEEE Transactions on Medical Imaging and of the IEEE Transactions on Pattern Analysis. Polina is currently on the editorial board of the Journal of Medical Image Analysis. She is a Fellow of the International Society for Medical Image Computing and Computer Assisted Interventions (MICCAI) and of the American Institute for Medical and Biological Engineering (AIMBE).

Learning to read xray: applications to heart failure monitoring

Abstract

We propose and demonstrate a novel approach to training image classification models based on large collections of images with limited labels. We take advantage of availability of radiology reports to construct joint multimodal embedding that serves as a basis for classification. We demonstrate the advantages of this approach in application to assessment of pulmonary edema severity in congestive heart failure that motivated the development of the method.

Christian Igel is a professor at DIKU, the Department of Computer Science at the University of Copenhagen. He studied Computer Science at the Technical University of Dortmund, Germany. In 2002, He received his Doctoral degree from the Faculty of Technology, Bielefeld University, Germany, and in 2010 my Habilitation degree from the Department of Electrical Engineering and Information Sciences, Ruhr-University Bochum, Germany. From 2003 to 2010, he was a Juniorprofessor for Optimization of Adaptive Systems at the Institut für Neuroinformatik, Ruhr-University Bochum. In October 2010, he was appointed professor with special duties in machine learning at DIKU. 

He has been a full professor at DIKU since December 2014. Christian is also the director of the SCIENCE AI Centre and a co-lead of the Pioneer Centre for Artificial Intelligence, Denmark. He is a Fellow of ELLIS, European Lab for Learning and Intelligent Systems.

His main research interests are support vector machines and other kernel-based methods, evolution strategies for single- and multi-objective optimization and reinforcement learning, PAC-Bayesian analysis of ensemble methods, and deep neural networks and stochastic neural networks.


Deep Learning and remote sensing for ecosystem monitoring
Abstract
Progress in remote sensing technology and machine learning algorithms enables scaling up the monitoring of ecosystems. This leads to new knowledge about their status and dynamics, which will be helpful in land degradation assessment (e.g., deforestation), in mitigating poverty (e.g., food security, agroforestry, wood products), and in managing climate change (e.g., carbon sequestration).

This talk will first present deep learning for the mapping of individual trees and forests. Tree crowns are segmented in satellite imagery using fully convolutional neural networks. This provides detailed measurements of the canopy area and of the distribution of trees within and outside forests. Allometric equations are applied to estimate the biomass (and thereby the stored carbon) of the individual trees. The talk will discuss some technical aspects of fitting and uncertainty quantification of allometric models.

Then it is shown how tree biomass can be directly inferred from LiDAR (laser imaging, detection, and ranging) measurements using 3D point cloud neural networks. This leads to highly accurate results without requiring a digital elevation model. In this context, we will discuss a general problem when using deep learning for least-squares regression, namely that the error residuals of the neural network do not necessarily vanish. This can lead to systematic errors that accumulate if we are interested in the total aggregated performance over many data points. We suggest addressing this issue as a default postprocessing step.