Data Science Research Seminar
The UM Data Science Research Seminar Series are monthly sessions organised by the Institute of Data Science in collaboration with different departments across UM with the aim to bring together data scientists from across Maastricht University to discuss breakthroughs and research topics related to Data Science.
This next session is organised in collaboration with the Department of Data Science and Knowledge Engineering on December 6, 2018 from 12:00 - 13:00.
Tapijn (Tapijnkazerne 20, 6221 ME, Maastricht), Building B. Room 0.0009
Event is free and open to everyone. Lunch will be provided. Please register by December 4, 2018.
Schedule
12.00 - 12.30
Talk by Christof Seiler, Assistant Professor, DKE
Title: A Multivariate Mixed Model for Single-Cells Biology
Abstract: In biology, the invention of mass cytometry made it possible for scientists to measure four times more proteins on single cells. This allowed immunologists to redefine most of the major subsets of white blood cells. Despite these technological advances, current statistical methods often collapse the rich single-cell data into summary statistics before proceeding with downstream analysis. In this talk, our aim is to promote statistical analyses on the unsummarized data. We will show that multivariate generative models are a valid alternative to univariate hypothesis testing. We will give an introduction to the multivariate Poisson log-normal distribution in the context of a recent pregnancy study. We will use Hamiltonian Monte Carlo for Bayesian inference. Our approach successfully reproduces key findings while extracting new information.
12.30 - 13.00
Talk by Siamak Mehrkanoon, Assistant Professor, DKE
Title: Learning from Partially Labeled Data
Abstract: Often obtaining labeled samples is a costly and time consuming process. Moreover, in today’s applications, evolving data streams are ubiquitous. Due to the complex underlying dynamics and non-stationary behavior of real-life data, the demand for adaptive learning mechanisms is increasing. In this context, designing models that can leverage rich labeled data in one domain and be applicable to a different but related domain is also highly desirable. Here, a so-called multi-class semi-supervised learning model (MSS-KSC) will be introduced. We further discuss two approaches to make the model scalable to large scale data. Next we briefly discuss its incremental version for on-line clustering/classification of time-evolving data. Last but not last, a brief overview of the recently proposed Regularized Semi-Paired Kernel Canonical Correlation Analysis (RSP-KCCA) formulation for domain adaptation problem will be provided.
For questions or concerns, please contact us via info-ids@maastrichtuniversity.nl.