
what's coming up
Please fill in this form if you're interested in joining as a student affiliate.
Stay updated on our latest events and news by joining our mailing list.
Interested in sharing your expertise? Apply to give a talk at our next seminar.
Speaker: Tim Reichelt
Time: 17:30-18:30
Levett Room, Wolfson College & Online (Teams link).
Biography:
I am an Encode AI Fellow in the Climate Processes group led by Prof Philip Stier. As part of the Encode fellowship, we are building a generative model for cloud structures with the primary goal is to constrain cloud climate feedbacks. Before starting the fellowship, I was conducting research on compression algorithms for climate and weather data funded by the Embed2Scale project.
I completed my PhD as part of the EPSRC Centre for Doctoral Training in Autonomous Intelligent Machines and Systems here in Oxford. In my research I developed novel Bayesian inference algorithms for models defined through probabilistic programs.
Abstract:
We are building a generative model for cloud structures conditioned on environmental conditions, integrating both high-resolution satellite data and reanalysis data. By being able to generate and reconstruct cloud structures under varying environmental conditions, the model will allow us to reduce uncertainties in future climate predictions and facilitate the detection and quantification of cloud feedbacks across the satellite record. This purely data-driven approach to constraining cloud feedbacks will be complementary to existing work that relies on physics-based models.
Speaker: Konstantin Hemker
Time: 17:30-18:30
Levett Room, Wolfson College & Online (Teams link).
Biography:
I grew up in the beautiful city of Hamburg in Germany and moved to the UK after high school for my undergraduate degree at the London School of Economics. About a year into my time at LSE, I realised that I was more interested in mathematics & statistics than the economics aspects of my degree, which eventually led me to self-educate myself in Computer Science on the side. I then did a Master’s in Computer Science at Imperial College London, where I focussed on various disciplines ranging from Cybersecurity to Natural Language Processing. I really enjoyed Natural Language Processing and Machine Learning more generally and started working as a Senior Data Scientist in BCG’s pharmaceutical practice. During my PhD at the University of Cambridge, I focused on representation learning for high-dimensional biomedical data and worked on projects with Flagship Pioneering and Microsoft Research.
Abstract:
This talk will cover methods used to learn multimodal and multiscale representations of human tissue samples from histology images, bulk sequencing, and spatial transcriptomics data derived from cancer biopsies. In particular, the increasing availability and resolution of spatially resolved sequencing on human tissue samples, such as Spatial Transcriptomics (ST), provides rich and spatially resolved molecular information to diagnose and analyse tumours beyond the morphological information routinely available to pathologists through Whole Slide Images. Complex morphological and molecular spatial information becoming available at scale requires building robust multimodal AI architectures that take advantage of such high-dimensional information. We will cover some of the methods we developed for this problem, such as efficiently learning from multi-resolution microscopy, some general multimodal architectures, and SSL for spatial transcriptomics.
Events held and scheduled by us
Watch recordings of past events on our YouTube channel.




























































































