Deep adversarial transition learning using cross-grafted generative stacks

Hou, Jinyong; Ding, Xuejie; Cranefield, Stephen; Deng, Jeremiah D.

doi:10.1016/j.neunet.2022.02.011

Cited by 6 publications

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The CDLM has higher adaptation accuracies for the scenarios with a seemingly large domain gaps, such as MNIST→MNISTM and Fashion→FashionM. For the 3D scenario, our model's performance is a little lower than PixelDA [30] and DATL [56] but outperforms all the other methods. In PixelDA, the input is not only source images but also depth image pairs.…”

Section: Quantitative Results On Unsupervised Domain Adaptationmentioning

confidence: 83%

“…For this scenario, only the labels of the source images were available during training. We choose DANN [37] as the baseline, and compare our model with several other state-ofthe-art domain adaptation methods, including Conditional Domain Adaptation Network (CDAN) [52], Pixel-level Domain Adaptation (PixelDA) [30], Unsupervised Imageto-Image translation (UNIT) [4], Cycle-Consistent Adversarial Domain Adaptation (CyCADA) [32], Generate to Adapt (GtA) [31], Transferable Prototypical Networks (TPN) [53], Domain Symmetric Networks (SymmNets-V2) [54], Instance Level Affinitybased Networks (ILA-DA) [55], and Deep Adversarial Transition Learning (DATL) [56].…”

Section: Quantitative Results On Unsupervised Domain Adaptationmentioning

confidence: 99%

See 1 more Smart Citation

Variational Transfer Learning using Cross-Domain Latent Modulation

Hou¹,

Deng²,

Cranefield³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

To successfully apply trained neural network models to new domains, powerful transfer learning solutions are essential. We propose to introduce a novel cross-domain latent modulation mechanism to a variational autoencoder framework so as to achieve effective transfer learning. Our key idea is to procure deep representations from one data domain and use it to influence the reparameterization of the latent variable of another domain.Specifically, deep representations of the source and target domains are first extracted by a unified inference model and aligned by employing gradient reversal. The learned deep representations are then cross-modulated to the latent encoding of the alternative domain, where consistency constraints are also applied. In the empirical validation that includes a number of transfer learning benchmark tasks for unsupervised domain adaptation and image-to-image translation, our model demonstrates competitive performance, which is also supported by evidence obtained from visualization.

show abstract

Section: Quantitative Results On Unsupervised Domain Adaptationmentioning

confidence: 83%