Semi-supervised Deep Representation Learning for Multi-View Problems

Abstract

While neural networks for learning representation of multi-view data have been previously proposed as one of the state-of-the-art multi-view dimension reduction techniques, how to make the representation discriminative with only a small amount of labeled data is not well-studied. We introduce a semisupervised neural network model, named Multi-view Discriminative Neural Network (MDNN), for multi-view problems. MDNN finds nonlinear view-specific mappings by projecting samples to a common feature space using multiple coupled deep networks. It is capable of leveraging both labeled and unlabeled data to project multi-view data so that samples from different classes are separated and those from the same class are clustered together. It also uses the inter-view correlation between views to exploit the available information in both the labeled and unlabeled data. Extensive experiments conducted on four datasets demonstrate the effectiveness of the proposed algorithm for multi-view semisupervised learning.

Multi-View Representation Learning with Manifold Smoothness

“…The representation they learned lacks class discriminativeness that is critical to the success of some tasks, i.e., classification and clustering. Recently, Noroozi et al (Noroozi et al 2018) proposed the first deep semi-supervised representation learning method Multi-view Discriminative Neural Network (MDNN) for multi-view problems. MDNN extends DCCA by considering both the correlation of all the data and the Linear Discriminant Analysis (LDA) objective function to maximize between-class separations and minimize within-class variations.…”

“…In the multi-view setting, examples are described with several disjoint sets of features. Most multi-view representation learning methods focus on two views (Akaho 2006;Andrew et al 2013;Hotelling 1936;Noroozi et al 2018;Wang et al 2015), we also discuss the two-view setting here. For a two-view problem, we can denote the data matrices on the two views as X 1 and X 2 , respectively.…”

“…In general, purely unsupervised representation learning methods could not utilize information in labeled data, and thus may not be sufficiently discriminative in downstream tasks such as classification or clustering. To tackle this, Noroozi et al (Noroozi et al 2018) recently proposed the first deep semi-supervised representation learning model Multi-view Discriminative Neural Network (MDNN) for multi-view problem, which could be viewed as a combination of Deep Linear Discriminant Analysis (DLDA) (Dorfer, Kelz, and Widmer 2016) and DCCA (Andrew et al 2013). However, MDNN did not incorporate the manifold information to maintain the smoothness over the data.…”

Multiscale Laplacian Learning

“…For WebKB data, we compare classification accuracy against recent methods such as semi-supervised multi-view deep discriminant representation learning (SMDDRL) [45], vertical ensemble co-training (VE-CoT) [51], auto-weighted multiple graph learning (AMGL) [78], multi-view learning with adaptive neighbors (MLAN) [75], deep canonically correlated autoencoder (DCCAE) [103], multi-view discriminative neural network (MDNN) [80], semi-supervised learning for multiple graphs by gradient flow (MGSC) [59], multidomain classification w/ domain selection (MCS) [22], multiview semi-supervised learning (FMSSL, FMSSL-K) [109], and semi-supervised multimodal deep learning framework (SMDLF) [25]. Our results are obtained using 105 labels, and using the classification accuracy metric.…”

Ανάπτυξη Μερικώς Επιβλεπόμενων Αλγορίθμων Μηχανικής Μάθησης

“…Ή ακόμη, να τους διαφωτίσει περί των προτιμήσεων που μπορεί να προέρχονται από την αντίθετη κατεύθυνση, από τα προβλήματα δηλαδή προς τους αλγορίθμους, όπως συμβαίνει εντόνως στα ακουστικά σήματα και τη χρήση των SVMs [742], [743]. Προς αντίστοιχη κατεύθυνση κινήθηκε η εργασία [142], η οποία συγκέντρωσε μαζικά αποτελέσματα σε μεγάλο πλήθος προβλημάτων με διάφορους αλγορίθμους, καθώς και η [768], [267], [279], [543], [762], [771], [772], [773].…”