Effective domain adaptation approaches for image and time series classification

Abstract

Deep learning has become widely used for training models on large datasets and has newlineshown significant progress across various applications. However, training such models newlineeffectively often requires a large amount of annotated data, which can be difficult to obtain newlinein certain domains. For example, in the medical field, data processing is challenging due newlineto privacy concerns. Similarly, in visual recognition tasks, labeling images is both timeconsuming and costly. Traditional machine learning assumes that training and testing data newlinecome from the same distribution, but this assumption often does not hold in real-world newlinescenarios. In practice, training and testing data may differ due to domain shifts caused by newlinefactors such as environmental conditions, lighting, viewing angles, equipment variations, newlinecamera types, background clutter, and other influences. These shifts can significantly newlineimpact the performance of standard machine learning models.The thesis focuses on developing effective domain adaptation techniques for image and newlinetime-series classification in an unsupervised domain adaptation setting. newline