Csáji, Balázs Csanád (SZTAKI)
Distribution-Free Guarantees for
Kernel Methods
Kernel methods are widely
used in statistics, machine learning (ML), signal processing, and related fields.
Their theoretical foundations are based on reproducing kernel Hilbert spaces.
Kernels are often interpreted in ML as they measure similarity between objects.
Typical kernel methods are, for example, various types of support vector
machines; moreover, it is often easy to kernelize any linear method (such as
ridge regression, LASSO or principal component analysis). In this talk, I give
a short introduction to learning with kernels, and then I present some recent
results providing distribution-free, non-asymptotic theoretical guarantees for
kernel methods.
Related papers
(references):
·
Csáji, B. Cs.; Kis, K. B.:
Distribution-Free Uncertainty Quantification for Kernel Methods by Gradient
Perturbations, Machine Learning, Springer, Vol. 108, 2019, pp. 1677–1699
·
Tamás, A.; Csáji, B.
Cs.: Exact Distribution-Free Hypothesis Tests for the Regression Function of
Binary Classification via Conditional Kernel Mean Embeddings, IEEE Control
Systems Letters, IEEE Press, Vol. 6, 2021, pp. 860-865.
·
Csáji, B. Cs.; Horváth,
B.: Nonparametric, Nonasymptotic Confidence Bands
with Paley-Wiener Kernels for Band-Limited Functions, IEEE Control Systems
Letters, IEEE Press, 2022 [in press]
The talk
is held in English!
Az előadás nyelve angol!
Date: Sep 6, Tuesday 4:15pm
Place: BME, Building „Q”,
Room QBF13