The field of hyperspectral image classification and reconstruction is moving towards the development of more efficient and effective models that can handle the challenges of high spectral dimensionality, complex spectral-spatial correlations, and limited training samples. Researchers are exploring the use of hybrid architectures that combine the strengths of convolutional neural networks and transformers to improve classification accuracy. Additionally, semi-supervised domain adaptation techniques are being investigated to bridge the gap between general and human-centered hyperspectral image datasets, which is crucial for medical applications. Another area of focus is the development of lightweight and effective deep learning models that can capture long-range dependencies in spectral and spatial dimensions. Noteworthy papers include CLAReSNet, which proposes a hybrid architecture that integrates multi-scale convolutional extraction with transformer-style attention, and SpectralAdapt, which introduces a semi-supervised domain adaptation framework that bridges the domain gap between general and human-centered HSI datasets. SpectralTrain is also a notable framework that enhances learning efficiency by integrating curriculum learning with principal component analysis-based spectral downsampling.