Seminar of Prof. E. Isufi: "Learning with Statistical Relations"

Info: Standard machine learning relies heavily on sample covariance matrices for network topology and dimensionality reduction (e.g., PCA). However, finite-data estimation errors often introduce spectrum uncertainties, making methods like PCA highly sensitive to small data perturbations.

In this talk, we explore CoVariance Neural Networks (VNNs)—an innovative architecture performing graph convolutions directly on sample covariance matrices to enhance stability and representation power. We will discuss:

Sparse VNNs: Overcoming high-dimensional, limited-data settings with theoretically grounded sparsification to boost efficiency and filter out spurious correlations.

SpatioTemporal VNNs: Processing multivariate time series via online covariance and temporal convolutions.

Fair VNNs: Leveraging equitable covariance estimates to promote fairness across underrepresented groups.

Event venue & nearby stays

Via Goffredo Duranti, 93, 93 Via Goffredo Duranti, Perugia, Italy

Concerts, fests, parties, meetups — all the happenings, one place.

Get the app →

Discover more events by tag