修正源標籤以改進半監督式域適應

Abstract

半監督式域適應涉及到學習使用少量的標記目標數據和許多未標記的目標數據，以及來自相關領域的標記源數據，以對未標記的目標數據進行分類。當前的半監督式域適應方法通常旨在通過特徵空間映射和偽標籤分配將目標數據與標記的源數據對齊。 然而，這種源導向的模型有時會將目標數據與錯誤類別的源數據對齊，從而降低分類的表現。 我們提出了一種新穎的域適應典範，可以調整源數據以匹配目標數據。 我們的核心思想是將源數據視為一種含有噪聲標記的理想目標數據。 我們提出了一個半監督式域適應模型，該模型借助從目標的角度設計的清理元件來動態清除源數據的噪聲標籤。 由於這種想法與現有的其他半監督式域適應方法背後的核心理念有很大的不同，因此，我們提出的模型可以很容易地與這些方法結合以提高它們的性能。 在兩種主流的半監督式域適應方法上的實驗結果表明，我們提出的模型有效地清除了源標籤內的噪聲，並在主流的數據集上得到優於這些方法的表現。

Semi-supervised domain adaptation (SSDA) involves learning to classify unseen target data with a few labeled data and many unlabeled target data, along with many labeled source data from a related domain. Current SSDA approaches typically aim at aligning the target data to the labeled source data with feature space mapping and pseudo-label assignment. Nevertheless, such a source-oriented model sometimes aligns the target data to source data of the wrong class, degrading the classification performance. We present a novel source-adaptive paradigm that adapts the source data to match the target data. Our key idea is to view the source data as a noisily-labeled version of the ideal target data. We propose an SSDA model that cleans up the label noise dynamically with the help of a robust cleaner component designed from the perspective of the target. Since this paradigm differs greatly from the core ideas behind existing SSDA approaches, our proposed model can be easily coupled with such approaches to improve their performance. Empirical results on two state-of-the-art SSDA approaches demonstrate that the proposed model effectively cleans up noise within the source labels and exhibits superior performance over those approaches across benchmark datasets.