Tighter Bounds for Structured Estimation

By Olivier Chapelle, Chuong B. Do, Quoc V. Le, Alex Smola, Choon Hui Teo. 2008

73 citations recorded by the source corpus.

Subfields: Statistics and Probability, Statistical Learning Theory

How researchers see this paper

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

In Statistics and Probability

In Statistical Learning Theory

Citations in the corpus

The landscape of empirical risk for nonconvex losses (2016, background): “First of all, several empirical studies [4, 21, 15] demonstrate an improved classification error using non-convex losses.”
Houdini: Fooling Deep Structured Prediction Models (2017, background): “Do et al. [9] generalized the notion of the ramp loss from binary classiﬁcation to structured prediction and proposed a tighter bound to the task loss than the structured hinge loss.”
Learning Syntax from Naturally-Occurring Bracketings (2021, methodology): “To overcome the challenge of learning from this kind of noisy data, we propose to train discriminative constituency parsers with structured ramp loss (Do et al., 2008), a technique previously adopted in machine translation (Gimpel and Smith, 2012).”
On the Convergence of the Concave-Convex Procedure (2009, methodology): “Other applications where CCCP has been used include sparse principal component analysis [27], transductive SVMs [11, 5, 28], feature selection in SVMs [22], structured estimation [10], missing data problems in Gaussian processes and SVMs [26], etc.”
Advanced Structured Prediction (2014, unknown): “The idea is to use structured ramp loss (Do et al., 2008) as a surrogate loss function, which is defined as follows: ̄̀ ramp(w,x,y) = max ŷ∈Y [ `(y, ŷ) +w>φ(x, ŷ) ] −max ỹ∈Y [ w>φ(x, ỹ) ] .”
Object Detection with Grammar Models (2011, background): “In this case L′ becomes a type of ramp loss [4, 6, 14].”
Learning structural SVMs with latent variables (2009, methodology): “Very recently the CCCP algorithm has also been applied to obtain tighter non-convex loss bounds on structured learning (Chapelle et al., 2008).”
"On the (Non-)existence of Convex, Calibrated Surrogate Losses for Ranking" (2012, methodology): “A first possibility is to turn to non-convex losses for ranking as in [10, 3], and to study the calibration of such losses.”
Ranking with ordered weighted pairwise classification (2009, background): “On the other hand, this upper bound can be very loose (Do et al., 2008), and there is little theoretical evidence on how to make the feature map.”