A note on Platt’s probabilistic outputs for support vector machines

By Hsuan-Tien Lin, Chih-Jen Lin, Ruby Chiu-Hsing Weng. Machine-mediated learning, 2007

1,003 citations recorded by the source corpus.

Subfields: Computational Theory and Mathematics, Control and Systems Engineering, Statistical and Nonlinear Physics

View Paper: https://www.semanticscholar.org/paper/a76a95c8766550c84be48efb70dc7391e84553df

How researchers see this paper

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

In Computational Theory and Mathematics

In Control and Systems Engineering

In Statistical and Nonlinear Physics

Citations in the corpus

A Human Activity Recognition System Using Skeleton Data from RGBD Sensors (2016, methodology): “Usually, SVMs do not provide output probabilities, but there exist some methods to extend SVM implementations and make them able to provide also this information [53, 54].”
Content-based visual search learned from social media (2012, methodology): “where a and b are two real-valued parameters optimized by solving a regularized maximum likelihood problem as described in [71].”
A Machine Learning Framework to Identify Students at Risk of Adverse Academic Outcomes (2015, methodology): “Support Vector Machines estimate the signed distance of a data point from the nearest hyperplane and Platt scaling can be used to convert these distances into probability estimates [19].”
Unsupervised ensemble ranking of terms in electronic health record notes based on their importance to patients (2017, methodology): “LibSVM transforms these distance values into probabilities by Platt scaling [56,57].”
Visual to Text: Survey of Image and Video Captioning (2019, methodology): “In order to match language description to out-of-vocabulary words, the algorithm utilizes the semantic distance in the WordNet according to the Lin’s measure [68], especially for objects and scenes.”
Multimodal information fusion for urban scene understanding (2016, methodology): “The maximization of (27) is done using Newton’s method [22], which only takes O(N ) time per iteration.”
Multi-class active learning for image classification (2009, methodology): “In order to obtain these estimates, we follow the approach proposed by [13], which is a modified version of Platt's method to extract probabilistic outputs from SVM [16].”
Looking at Vehicles on the Road: A Survey of Vision-Based Vehicle Detection, Tracking, and Behavior Analysis (2013, methodology): “Competing discriminative methods converge to a global optimum over the training set, which provides nice properties for further analysis, such as fitting posterior probabilities [188].”
Estimating accurate multi-class probabilities with support vector machines (2005, methodology): “But, in a recent note [13], it is shown that there are two problems in the pseudo-code provided in [18].”
deepNF: deep network fusion for protein function prediction (2017, methodology): “We use Platt calibration [4] to convert GeneMANIAs scores into probability score before evaluation, as suggested in the Mashup paper [5].”
Concept-Based Video Retrieval (2009, methodology): “[134] proposed a computational improvement over Platt’s algorithm.”
Tree Kernels for Semantic Role Labeling (2008, methodology): “3 We actually implemented the pseudo-code proposed in Lin, Lin, and Weng (2003) which, with respect to Platt’s original formulation, is theoretically demonstrated to converge and avoids some numerical difficulties that may arise.”
Temporal Localization of Actions with Actoms (2013, neutral): “In order to return probability estimates, we ﬁt a sigmoid function to the decision function f learned by the SVM [68, 69].”
Non-Parametric Calibration for Classification (2019, background): “For example, Pereyra et al. (2017) propose to penalize low-entropy output distributions, Kumar et al. (2018) (Murphy, 1973, DeGroot and Fienberg, 1983) and approaches to provide uncertainty estimates for non-probabilistic binary classiﬁers (Platt, 1999, Lin et al., 2007, Zadrozny and Elkan, 2002).”
LIBSVM: A library for support vector machines (2011, neutral): “Therefore, the complexity of Algorithm 1 is Several works have studied the number of iterations of decomposition methods; see, for example, List and Simon (2007).”
Feature selection via sensitivity analysis of SVM probabilistic outputs (2008, methodology): “Since (8) is an unconstrained convex optimization problem in two variables, its numerical determination is straight forward (Lin et al. 2003).”