U
Umut Simsekli
Researcher at French Institute for Research in Computer Science and Automation
Publications - 125
Citations - 1970
Umut Simsekli is an academic researcher from French Institute for Research in Computer Science and Automation. The author has contributed to research in topics: Stochastic gradient descent & Computer science. The author has an hindex of 20, co-authored 110 publications receiving 1409 citations. Previous affiliations of Umut Simsekli include Télécom ParisTech & Mitsubishi Electric.
Papers
More filters
Proceedings Article
Generalized Sliced Wasserstein Distances
TL;DR: In this paper, the generalized sliced-Wasserstein distance (GSW) distance was proposed for probability measures, which can be extended to a maximum GSW distance (max-GSW distance).
Proceedings Article
Generalised Coupled Tensor Factorisation
TL;DR: Algorithms for generalised tensor factorisation are derived that can compute arbitrary factorisations in a message passing framework, derived for a broad class of exponential family distributions including special cases such as Tweedie's distributions corresponding to β-divergences.
Posted Content
A Tail-Index Analysis of Stochastic Gradient Noise in Deep Neural Networks
TL;DR: In this paper, the gradient noise in SGD is considered in a more general context and the generalized central limit theorem (GCLT) is invoked to analyze SGD as an SDE driven by a Levy motion.
Proceedings Article
Sliced-Wasserstein Flows: Nonparametric Generative Modeling via Optimal Transport and Diffusions
TL;DR: In this article, a nonparametric IGM algorithm with explicit theoretical guarantees is proposed, which is based on a functional optimization problem, which aims at finding a measure that is close to the data distribution as much as possible and also expressive enough for generative modeling purposes.
Journal ArticleDOI
Real-time recognition of percussive sounds by a model-based method
TL;DR: A model-based algorithm for detection of percussive events is introduced and the results indicate that the approach is promising and applicable in design and development of interactive musical systems.