Efficient Learning of Domain-invariant Image Representations

Abstract

We present an algorithm that learns representations which explicitly compensate for domain mismatch and which can be efficiently realized as linear classifiers. Specifically, we form a linear transformation that maps features from the target (test) domain to the source (training) domain as part of training the classifier. We optimize both the transformation and classifier parameters jointly, and introduce an efficient cost function based on misclassification loss. Our method combines several features previously unavailable in a single algorithm: multi-class adaptation through representation learning, ability to map across heterogeneous feature spaces, and scalability to large datasets. We present experiments on several image datasets that demonstrate improved accuracy and computational advantages compared to previous approaches.

References in corpus (2)

• Frustratingly Easy Domain Adaptation
• Learning with Augmented Features for Heterogeneous Domain Adaptation

Cited by in corpus (20)

• DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition
• An introduction to domain adaptation and transfer learning
• LSDA: Large Scale Detection Through Adaptation
• Personalized Classifier for Food Image Recognition
• Kernel Manifold Alignment
• Optimal Bayesian Transfer Learning
• Hierarchically Learned View-Invariant Representations for Cross-View Action Recognition
• Score Function Features for Discriminative Learning: Matrix and Tensor Framework
• SPIGAN: Privileged Adversarial Learning from Simulation
• Heterogeneous Domain Adaptation with Adversarial Neural Representation Learning: Experiments on E-Commerce and Cybersecurity
• Robust Importance Sampling for Error Estimation in the Context of Optimal Bayesian Transfer Learning
• On Regularization Parameter Estimation under Covariate Shift
• Weakly Supervised Open-set Domain Adaptation by Dual-domain Collaboration
• Unsupervised Domain Adaptation with Progressive Domain Augmentation
• Joint cross-domain classification and subspace learning for unsupervised adaptation
• Heterogeneous Domain Adaptation via Soft Transfer Network
• SALT: Subspace Alignment as an Auxiliary Learning Task for Domain Adaptation
• Multi-source Heterogeneous Domain Adaptation with Conditional Weighting Adversarial Network
• Open Set Domain Adaptation for Image and Action Recognition
• Joint Information Preservation for Heterogeneous Domain Adaptation