Learning to rank models are broadly applied in ad hoc retrieval for scoring and sorting documents based on their relevance to textual queries. The generalizability of the trained model in the learning to rank approach, however, can have an impact on the retrieval performance, particularly when data includes noise and outliers, or is incorrectly collected or measured. In this paper, we introduce a Self-Distilled Learning to Rank (SDLR) framework for ad hoc retrieval, and analyze its performance over a range of retrieval datasets and also in the presence of features’ noise. SDLR assigns a confidence weight to each training sample, aiming at reducing the impact of noisy and outlier data in the training process. The confidence weight is approximated based on the feature’s distributions derived from the values observed for the features of the documents labeled for a query in a listwise training sample. SDLR includes a distillation process that facilitates passing on the underlying patterns in assigning confidence weights from the teacher model to the student one. We empirically illustrate that SDLR outperforms state-of-the-art learning to rank models in ad hoc retrieval. We thoroughly investigate the SDLR performance in different settings including when no distillation strategy is applied; when different portion of data are used for training the teacher and the student models, and when both teacher and student models are trained over identical data. We show that SDLR is more effective when training data are split between a teacher and a student model. We also show that SDLR’s performance is robust when data features are noisy.

The growing deployment of renewable energy systems calls for accurate parameter estimation and compact modeling of grid-connected inverter systems (GCISs). This paper presents a framework for parameter estimation and physics-structured model compression using Physics-Structured-Informed Neural Networks (Ψ-NNs). A teacher physics-informed neural network (PINN) estimates LC-filter parameters by combining measurements with governing equations. A regularized student learns the teacher mapping through physics-informed distillation. Hierarchical clustering identifies shared weight magnitudes, enabling structured reconstruction and fine-tuning. The framework is evaluated on mathematical-model and MATLAB/Simulink datasets. The teacher achieves a mean relative parameter error below $5\%$ across all cases, outperforming the extended Kalman filter, remaining competitive with least squares on mathematical-model data, and outperforming both baselines on all MATLAB/Simulink records. The reconstructed networks retain only 23–78 trainable base coefficients (0.15–0.50% of the student’s 15,600 parameters): 23–55 (0.15–0.35%) on the mathematical-model records and 62–78 (0.40–0.50%) on the MATLAB/Simulink records. These ratios count tied magnitudes only; after cluster assignments and signs are stored, the packed footprint is about 19–22% of the student model. The representation is value-tying rather than sparsity, so dense inference is not automatically faster. Reconstruction reduces the student data residual by nearly two orders of magnitude on degraded mathematical-model data, while remaining comparable on MATLAB/Simulink data except under the strongest combined artifacts. These results demonstrate the framework’s potential to combine robust parameter estimation with compact representations of GCIS dynamics, revealing a compression–accuracy trade-off under severe measurement degradation.