Learning to Rank Using Privileged Information

By Viktoriia Sharmanska, Novi Quadrianto, Christoph H. Lampert. 2013

199 citations recorded by the source corpus.

Many computer vision problems have an asymmetric distribution of information between training and test time. In this work, we study the case where we are given additional information about the training data, which however will not be available at test time. This situation is called learning using privileged information (LUPI). We introduce two maximum-margin techniques that are able to make use of this additional source of information, and we show that the framework is applicable to several scenarios that have been studied in computer vision before. Experiments with attributes, bounding boxes, image tags and rationales as additional information in object classification show promising results.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Information Retrieval & Search

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

In Computer Vision and Pattern Recognition

In Information Retrieval & Search

Citations in the corpus

Webly Supervised Joint Embedding for Cross-Modal Image-Text Retrieval (2018, methodology): “ for representing tags since tags do not have any semantic context as in the sentences. Our task can also be considered from the perspective of learning with side or privileged information strategies [45, 51], as in our case an additional tag modality is available at training time and we would like ”
Unifying distillation and privileged information (2015, methodology): “After distillation, we can use the simpler f s ∈ F s for faster prediction at test time.”
End-to-End Learning of Driving Models from Large-Scale Video Datasets (2016, background): “We propose to exploit privileged learning [24, 22, 8] to learn a driving policy that exploits both task loss and available side losses.”
Fast Generalized Distillation for Semi-Supervised Domain Adaptation (2017, methodology): “(Sharmanska, Quadrianto, and Lampert 2013) proposed a Rank Transfer method that uses attributes, annotator rationales, object bounding boxes, and textual descriptions as the privileged information for object recognition.”
Fashion IQ: A New Dataset Towards Retrieving Images by Natural Language Feedback (2021, background): “Learning with privileged information that is available at training time but not at test time is a popular machine learning paradigm [61], with many applications in computer vision [49, 25].”
Information Bottleneck Learning Using Privileged Information for Visual Recognition (2016, methodology): “The same idea has been used in the learning to rank approach introduced in [33], where it is shown that different types of privileged information, such as bounding boxes, attributes, text, and annotator rationales [7] can be used for learning a better classifier for object recognition.”
Sidekick Policy Learning for Active Visual Exploration (2018, background): “In learning with privileged information, an “expert” provides the student with training data having extra information (unavailable during testing) [61,53,37].”
Classifier learning with hidden information (2015, methodology): “Following the same experimental setting in [3], we perform 45 pair-wise classifications among a total of 10 classes, namely, chimpanzee, giant panda, leopard, persian cat, pig, hippopotamus, humpback whale, raccoon, rat and seal from the Animals with Attributes (AwA) dataset [24].”
Adaptive SVM+: Learning with Privileged Information for Domain Adaptation (2017, methodology): “We followed the same experimental procedure with [24, 29, 37] in which out of the 50 animal classes, we used the 10 testing classes for a total of 6,180 images.”
Privileged Information for Modeling Affect In The Wild (2021, background): “LUPI provides the means to transfer knowledge from all the available modalities to a machine learning model that makes predictions using only a subset of these modalities [40], [41].”
Object Localization based on Structural SVM using Privileged Information (2014, background): “Recently, [21] presented the potential benefit of SVM+ in object recognition task.”
Supervised mid-level features for word image representation (2014, background): “In that sense, our work can also be related to works on learning with privileged information [37, 33, 10], where the information available at training time to construct the representations (in our case, character bounding boxes) is not available at test time.”
From Dependence to Causation (2016, background): “…(Pechyony and Vapnik, 2010), equivalence analysis to weighted learning (Lapin et al., 2014), and multiple applications that include ranking (Sharmanska et al., 2013), computer vision (Sharmanska et al., 2014; LopezPaz et al., 2014), clustering (Feyereisl and Aickelin, 2012), metric…”
Simple and Efficient Learning using Privileged Information (2016, methodology): “The SVM+ at-tracts much attention recently [Vapnik and Izmailov, 2015; Lopez-Paz et al. , 2015; Lapin et al. , 2014], and has been successfully applied to different applications [Niu et al. , 2016; Sharmanska et al. , 2013; Fouad et al. , 2012].”
Cross-Domain Image Retrieval with a Dual Attribute-Aware Ranking Network (2015, background): “[36] explored attributes and rationales as a form of privileged information [43], considering a learning to rank framework.”
Learning with Side Information through Modality Hallucination (2016, background): “The theoretical framework was explored in [34] and a max-margin framework for learning with side-information in the form of bounding boxes, image tags, and attributes was examined in [26], while Shrivastava and Gupta [27] showed how surface normals at training time could produce detection improvemen”
Information Bottleneck Domain Adaptation with Privileged Information for Visual Recognition (2016, methodology): “One LUPI implementation is the SVM+ [45, 29], later extended to a learning to rank approach in [38], where it is shown that different types of auxiliary information, such as bounding boxes, attributes, and text can be used for learning a better classifier for object recognition.”