The field of data-driven modeling and non-intrusive load monitoring is moving towards the development of more accurate and efficient methods for forecasting and analyzing complex systems. Recent research has focused on incorporating prior knowledge and uncertainty into deep learning models, as well as improving the robustness and generalizability of these models. Notable advancements include the use of wavelet-based disentangled adaptive normalization for non-stationary time series forecasting and the development of transformer-based architectures for non-intrusive load monitoring. Additionally, weakly supervised approaches have shown promise in appliance localization and pattern detection. Noteworthy papers include: Initial Model Incorporation for Deep Learning FWI, which demonstrates the benefits of denormalization over pretraining for incorporating prior knowledge into neural networks. Wavelet-based Disentangled Adaptive Normalization for Non-stationary Times Series Forecasting, which proposes a model-agnostic framework for addressing non-stationarity in time series forecasting. NILMFormer, which introduces a transformer-based architecture for non-intrusive load monitoring that incorporates a novel positional encoding scheme. Few Labels are all you need, which presents a weakly supervised framework for appliance localization in smart-meter series that requires only presence information to be trained. History-Aware Neural Operator, which develops an autoregressive model for data-driven modeling of path-dependent inelastic materials that overcomes self-consistency issues and mitigates sensitivity to initial hidden states.