• "Complex and simple multidimensional real-life datasets"
  • "Principal graphs approximating single cell data"
  • "Mathematical theory of neural network correctors"

Complex and Simple Models of Multidimensional Data : from graphs to neural networks

Online half-day workshop: 1 December 2021, 2pm-5pm CET time

Branching Principal Components

Real-life datasets are characterized by a variety of geometries, topologies, ambient and intrinsic dimensionalities. In order to deal with this variety and complexity, we need to develop appropriate theoretical models of the data, able to capture their properties. Graph-based (such as principal graphs based on application of topological grammars) and neural network-based (such as non-linear autoencoders) methods have become popular recently in machine learning field and in practical applications such as the analysis of single cell molecular data. Simpler data models can be easier to manage and interpret but can miss important aspects of geometrical multidimensional data organization, while more complex models might be difficult to train and avoid overfitting. Any practical data analysis should determine a match between the data and the data model complexities. The purpose of this workshop is to collect early career and experienced researchers interested in the questions related to dealing with multidimensional data and data models.

Program of the workshop (CET time zone):

  • 14:00-14:40: Cees van Leeuwen
    University of Leuven
    Nearly two decades to nearly complete a research program:
    Past, present, and future of the Adaptive Rewiring project.

  • 14:40-15:20: Ivan Tyukin
    AIDAM Centre Leicester, UK; Lobachevsky University, Nizhniy Novgorod, Russia.
    The mathematics of learning from high-dimensional low-sample data with
    small neural networks.

  • 10' Coffee break.

  • 15:30-16:10: Luca Pinello
    Harvard Medical School; Massachusetts General Hospital
    SIMBA: SIngle-cell eMBedding Along with features based on graph embedding.

  • 16:10-16:50: Andrei Zinovyev
    Institut Curie, Paris Artificial Research Institute
    Elastic principal graphs to analyze data point clouds with complex geometry.

  • 16:50-17:20: Informal discussion.

A recording of the workshop is available here