A new arXiv preprint proposes using transfer learning to estimate socioeconomic conditions among forcibly displaced populations. The authors note that inclusive household surveys provide valuable welfare benchmarks but are costly and only conducted periodically, motivating cheaper modeling approaches. The work appears as both a new cs.LG submission and a cs.AI cross-list.
A new arXiv preprint proposes selecting source datasets for time series classification by measuring similarity through shapelets, the discriminative subsequences that characterize time series patterns. The authors argue that transfer learning helps overcome limited labeled data in deep learning, but its usefulness hinges on picking appropriate source datasets. Their method is presented as an alternative to conventional transferability estimation, which the paper describes as computationally expensive.
A new arXiv preprint introduces QSTAR, a method that decides when a quantum component should be used within a transfer-learning pipeline instead of always relying on a fixed variational quantum classifier. The authors argue that common evaluations of quantum transfer learning obscure this question, and their approach adds adaptive routing to select the quantum branch only when it contributes. The work is a research contribution and has not been peer-reviewed or released as a product.
A new arXiv paper investigates how pretrained language models can be adapted for univariate time-series forecasting using parameter-efficient transfer learning. The authors focus on identifying which design decisions matter most for effective transfer between text and numerical sequences. The work is a cross-listed submission to arXiv's machine learning category.
A research paper examines whether typing-pattern biometrics can be adapted for cross-device user authentication. Keystroke dynamics already performs well for single-device logins, but transferring the model between different keyboards and devices remains a challenge. The work targets applications such as fraud prevention.
A new preprint on arXiv describes a method that applies iterative sequential knowledge transfer to few-shot multiobjective multitask optimization. The work targets the challenge of sharing useful information across related optimization tasks when only limited data is available. It focuses on improving transfer mechanisms, a key bottleneck in multitask optimization research.