Contemporary sciences abound in various complex data types (beyond the classical vector description) including graphs, rankings, manifolds, time series, sets, probability measures, functions, or persistence diagrams, with numerous successful applications for instance in healthcare, bioinformatics, climate research, cosmology, manufacturing, computer vision, finance, economics, materials sciences, chemistry, or geosciences. This inherent heterogeneity calls for pushing the boundaries of statistical and machine learning methods to handle efficiently non-standard data by (i) designing computationally tractable, scalable and principled approaches, and understanding their statistical-computational trade-offs, (ii) leveraging the intrinsic data structure (e.g., invariant or multi-scale nature, using different notions of similarity), (iii) representing and quantifying the uncertainty of predictive models, and (iv) investigating the efficiency of the developed techniques in novel applications, with widespread social impact.
The goal of this workshop is:
1. to bring together researchers from mathematics, statistics, data science, and computer science to discuss the latest techniques, theoretical considerations, and applications for structured data, and
2. to explore future research directions and application avenues, accelerating the advances of the field.
We encourage contributed talks and posters in a variety of topics and methods both from theoretical and application point of view, including but not limited to:
- learning on graphs and networks,
- functional, topological, and object-oriented data analysis, shape analysis,
- analysis of distributional data, Bayes spaces, compositional data analysis,
- geometric learning, metric and kernel-based approaches, optimal transport,
- Bayesian inference, Gaussian processes,
- group-theoretical methods, leveraging invariances,
- information theoretical measures, hypothesis tests,
- uncertainty quantification on multivariate and non-Euclidean data,
- structured prediction, multi-task learning, surrogate losses,
- quantile measures and statistical depths,
- large-scale approximations, computational-statistical trade-offs,
- novel applications for learning on structured data.
Organisers:
- principal workshop organizer: Zoltan Szabo
- co-organizers: Alessandra Menafoglio, David Ginsbourger, Florence d’Alché-Buc, Judith Rousseau, Neil Lawrence.
Find out more about the workshop here.