Unsupervised Domain Adaptation via Stacked Convolutional Autoencoder

Unsupervised domain adaptation involves knowledge transfer from a labeled source to unlabeled target domains assist learning tasks. A critical aspect of unsupervised is the more transferable and distinct feature representations different domains. Although previous investigations, using, for example, CNN-based auto-encoder-based methods, have produced remarkable results in adaptation, there are still two main problems that occur with these methods. The first training problem deep neural networks; some optimization methods ineffective when applied networks second arises redundancy image data performance degradation adaptation. To address problems, this paper, we propose an method stacked convolutional sparse autoencoder, which based on performing layer projection original obtain higher-level More specifically, network, lower layers generate discriminative features whose kernels learned via autoencoder. reconstruction independent component analysis algorithm was introduced perform individual input data. Experiments undertaken demonstrated superior classification up 89.3% terms accuracy compared several state-of-the-art such as SSRLDA TLMRA.