DIKU Bits: How to learn from a confused teacher: Learning from uncertain data
Speaker: Aasa Feragen, associate professor in the Image Section
The cheapest and least invasive medical imaging technologies are typically also the noisiest, meaning that extracted information such as spatial location, shape, and condition of organs have a high degree of uncertainty associated with them. When, in clinical research, we want to learn from the extracted information using statistics or machine learning, the common approach is to ignore this uncertainty. In this talk, we will present you with diffusion MRI, an imaging modality used to learn about structural brain connectivity in live human beings. We discuss how we propagate the uncertainty in low-level image information to more advanced models of brain connectivity, and how this gives a more truthful representation of the extracted information. Next, we discuss the need to adapt statistics and machine learning in order to actually learn from this data, now that it no longer consists of data points -- but rather of data point distributions. The work is performed in collaboration with Anton Mallasto and Tom Dela Haije.
The talk will be conducted in English
New lecture series at DIKU
In this new lecture series you can get a closer look at the research conducted at DIKU, be motivated you to follow your interest in computer science and be inspired when choosing a subject for your bachelors project. The lectures are addressed to DIKU bachelor students - however, everyone is welcome.