Bregman Divergences and Surrogates for Learning

By Richard Nock, Frank Nielsen. 2009

72 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Private federated learning on vertically partitioned data via entity resolution and additively homomorphic encryption (2017, methodology): “Nevertheless, square losses are commonly used for classification [Nock and Nielsen, 2009] and that is essentially what we have obtained.”
Entity Resolution and Federated Learning get a Federated Resolution (2018, background): “Popular examples of RSPLs include the square, logistic and Matsushita losses [21], the square loss also being a Taylor loss.”
Ieee Transactions on Pattern Analysis and Machine Intelligence Classification with Noisy Labels by Importance Reweighting (unknown): “In this paper, we exploit the ratio matching approach that employs the Bregman divergence [51] (KLIEP [52]) to estimate the conditional probability distribution P Dρ ( ˆ Y |X).”
Loss factorization, weakly supervised learning and label noise robustness (2016, methodology): “The Factorization Theorem 3 generalizes [Patrini et al., 2014, Lemma 1] that works for symmetric proper losses ( spl s), e.g. logistic, square and Matsushita losses.”
Weakly supervised learning via statistical sufficiency (2016, background): “The class of symmetric proper losses (spls) [Nock and Nielsen, 2009], e.g. logistic, square and Matsushita losses, satisfies the linear-odd condition.”
Learning Classifiers with Fenchel-Young Losses: Generalized Entropies, Margins, and Algorithms (2018, background): “Since the best surrogate loss often depends on the data (Nock and Nielsen, 2009), tuning α typically improves accuracy, as we confirm in §6.”