Smart-city systems often face a data imbalance: labeled examples are scarce and expensive to obtain, while unlabeled observations are abundant. Semi-supervised learning (SSL) is designed for exactly this setting, using both types of data to improve classification when supervision is limited.
The paper, posted on arXiv, proposes a multi-objective evolutionary approach to SSL that is explicitly resource-aware. By framing the learning problem as a multi-objective evolutionary search, the method aims to produce classifiers that are suitable for deployment in resource-constrained smart-city environments.
Because the abstract is truncated, the available excerpt does not include experimental results or comparisons. The contribution described is the combination of evolutionary multi-objective optimization with SSL for resource-aware classification, a direction that addresses a practical bottleneck in smart-city analytics.