Learn to Transfer Statistical Models from and to Populations
Talk, UCA, AI GT, UCA, France
A short description:
- Motivations and applications
- Background
- Problem formulation
- Constraints, hypothesises, and properties
- Experimental results
- Open problems
Talks and presentations
Learning and inferring on Grassmannian Manifolds: With Applications on Schizophrenia Data
Talk, UCA, France
A short description:
- Motivations and applications
- Linear and Geodesic regression
- Background: Stiefel and Grassmann manifolds
- Problem formulation : The regression model
- Monte Carlo Sampling on Grassmannian manifolds (MCG)
- Experimental results
- Discussion
Regression on Grassmannian Manifolds
Conference talk, PR International Conference on Artificial Intelligence, Jakarta, Indonesia
A short description:
- Problem formulations: Example of Euclidean least-squares estimate
- Problem formulations: Example of geodesic least-squares estimate
- Problem formulation on the Grassmannian: regression model
- The geometric structure
- The Bayezian optimization with Hamiltonian dynamics
- Applications
Tutorial: Transfer Learning on Riemannian Manifolds
Tutorial, University of Sousse, Sousse, Tunisia
A short description:
- Study the geometry structure of some Riemannian manifolds
- Definitions and characterisation of transfer learning
- Statistical models and the geometric structure of the underlying space
- Choice of appropriate metrics
- Transfer learning algorithms of statistical models with parallel transport.
- Applications and Extensions
Bayesian Optimization for Classifying Probability Density Functions
Talk, Politecnico di Torino, Turin, Italy
A shhort description:
- Data representation leads to space of representations
- PDFs and densities to represent data
- Some limitations to extend ML techniques to non-flat spaces
- Examples from functional data
- A new Bayesian optimization method
- Examples from applications
Learn, predict and infer on Probability Density Functions
Talk, Technical University of Munich, AI in Medicine, Munich, Germany
A short description:
- Introduce a new manifold structure for the space of PDFs.
- Apply constrained GP to learn and infer on PDF.
- Extend to CDFs.
- Use spherical HMC sampling to solve the posterior probability on coefficients.
- A framework that can be generalized to other manifolds.
- Some applications in medical