Canonical Correlation Analysis: An Overview with Application to Learning Methods

By David Roi Hardoon, Sándor Szedmak, John Shawe-Taylor. 2004

3,365 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Statistics and Probability, Representation & Self-Supervised Learning, Signal Processing

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 Statistics and Probability

In Representation & Self-Supervised Learning

In Signal Processing

Citations in the corpus

Self-Supervised Learning of Visual Features through Embedding Images into Text Topic Spaces (2017, neutral): “87 CCA [15, 29] 19.”
Fine-Grained Image Classification via Combining Vision and Language (2017, methodology): “tion method is the canonical correlation analysis (CCA) [20], which learns a subspace to maximize the correlation among data of different media types, and is widely used for modeling multi-modal data [21,22,23]. Zhai et al. [24] propose to learn projection functions by the metric learning, and this ”
Framing Image Description as a Ranking Task: Data, Models and Evaluation Metrics (2013, methodology): “Most of the systems we present are based on a technique called Kernel Canonical Correlation Analysis (Bach & Jordan, 2002; Hardoon et al., 2004).”
Beyond CCA: Moment Matching for Multi-View Models (2016, neutral): “Examples of such data include, for instance, representation of some text in two languages (e.g., Vinokourov et al., 2002) or images aligned with text data (e.g., Hardoon et al., 2004; Gong et al., 2014).”
Learning Deep Structure-Preserving Image-Text Embeddings (2015, methodology): “Kernel CCA [17] is an extension of CCA in which maximally correlated nonlinear projections, restricted to reproducing kernel Hilbert spaces with corresponding kernels, are found.”
Multimodal Compact Bilinear Pooling for Visual Question Answering and Visual Grounding (2016, methodology): “Some of such embeddings are based on Canonical Correlation Analysis (Hardoon et al., 2004) e.g. (Gong et al., 2014; Klein et al., 2015; Plummer et al., 2015), linear models with ranking loss (Frome et al., 2013; Karpathy and Fei-Fei, 2015; Socher et al., 2014; Weston et al., 2011) or non-linear deep”
Multimodal Deep Learning (2011, methodology): “In particular, we suggest using canonical correlation analysis (CCA) (Hardoon et al., 2004), which finds linear transformations of audio and video data, to form a shared representation.”
A Multi-View Embedding Space for Modeling Internet Images, Tags, and Their Semantics (2012, methodology): “We have started with the two-view visual-textual CCA model popular in several recent works (Gong and Lazebnik, 2011; Hardoon et al., 2004; Hwang and Grauman, 2010, 2011; Rasiwasia et al., 2010) and shown that its performance can be significantly improved by adding a third view based on semantic grou”
A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI (2019, methodology): “sed (SVD) covariance matrix. The PC with the highest eigenvalue is roughly the most informative feature. Many kernel functions have been introduced, including the Canonical Correlation Analysis (CCA) [112]. CCA provides the set of features that transforms the original variables to the pairs of canon”
Transductive Multi-View Zero-Shot Learning (2015, methodology): “Thus, with our framework, we can now infer attributes for any test instance via the learned embedding space Γ as ˆ y Au .”
Deep Multimodal Feature Analysis for Action Recognition in RGB+D Videos (2016, methodology): “As a baseline to the proposed method to perform the shared-specific analysis of the RGB+D inputs, we combined canonical correlation analysis (CCA) [19, 16] and reconstruction independent component analysis (RICA) [28], to extract correlated and independent components of input features.”
Learning Two-Branch Neural Networks for Image-Text Matching Tasks (2017, methodology): “One of the most popular baselines for image-text embedding is Canonical Correlation Analysis (CCA), which finds linear projections that maximize the correlation between projected vectors from the two views [21], [22].”
Weakly-Supervised Alignment of Video with Text (2015, background): “[17], who learn a kernelized version of CCA to rank images given text.”
Automatic Description Generation from Images: A Survey of Models, Datasets, and Evaluation Measures (2016, methodology): “The authors show that their model outperforms previously used KCCA approaches such as Hodosh and Hockenmaier (2013). Karpathy et al. (2014) extend the previous multi-modal embeddings model. Rather than directly mapping entire images and sentences into a common embedding space, their model embeds mor”
Investigating Multilingual NMT Representations at Scale (2019, methodology): “Uses Canonical Correlation Analysis (CCA) (Hardoon et al., 2004) to linearly transform l′1 and l ′ 2 to be as aligned as possi- ble, i.e., CCA computes l̃1 = W1l′1 and l̃2 = W2l ′ 2 to maximize the correlations ρ̄ = {ρ1, ..., ρmin(d′1,d′2)} between the new subspaces.”
Unsupervised Model Selection for Variational Disentangled Representation Learning (2019, background): “Other related approaches worth mentioning are the Canonical Correlation Analysis (CCA) and its modiﬁcations (Hardoon et al., 2004; Raghu et al., 2017; Morcos et al., 2018).”
Beyond Instance-Level Image Retrieval: Leveraging Captions to Learn a Global Visual Representation for Semantic Retrieval (2017, methodology): “As an alternative to CCA, previous work has learned the joint embedding with a ranking loss.”
Cross-View Image Geolocalization (2013, methodology): “Kernelized Canonical Correlation Analysis (KCCA): KCCA is a tool to learn the basis along the direction where features in different views are maximally correlated: max wx =0,wy =0 w x Σxywy√ w x Σxxwx √ w y Σyywy (2) where Σxx and Σyy represent the covariance matrices for ground-level feature and ae”