Abstract

This ICML 2016 workshop explores the intersection between geometry and machine learning. Topics of interest include statistics and optimization on manifolds, manifold learning, computations in stratified spaces and more. The workshop consists mostly of invited talks, but we also accept poster contributions.


Update: The workshop is now over. We thank the speakers for their excellent work, the poster presenters for creating such a lively session, and the audience for their engaging discussions. This was awesome!

Topic

Many Machine ­Learning (ML) problems are fundamentally geometric in nature, e.g.:

  • finding optimal subspaces can be recast as finding point estimates on the Grassmannian; 
  • multi-metric learning can be recast as the learning of a Riemannian tensor; 
  • symmetric-positive definite matrices form a nonlinear space. 

In spite of the above, however, most practitioners often do not treat these problems as such, thus missing many potential benefits; in other words, ignoring the underlying geometry often leads to sub­-optimal models and results that can be significantly improved. Furthermore, many real-­world problems that involve both geometry and statistical learning are hardly considered by the ML community due to lack of awareness or familiarity with the appropriate tools. This workshop will be based on a diverse list of world experts who agreed to give invited talks on several different aspects of the intersection of geometry and ML. We hope this will raise more awareness within the mainstream ML community to the important role of geometry in ML as well as to both the challenges and the advantages that this intersection between the fields gives rise to.  

Workshop format

This is a one-day workshop on June 23, 2016. The workshop will mostly focus on invited talks to promote geometric methods to interested ML researchers. The workshop will be open for poster submissions which will be non-proceedings; see our CfP for details. The workshop will accept abstracts describing both published and unpublished work.

Funding

We are grateful for funding from the Danish Council for Independent Research (Natural Sciences) to cover some workshop expenses. We would not have had such an excellent line-up of speakers without this funding.