When mathematical biology models are used to make quantitative predictions for clinical or industrial use, it is important that these predictions come with a reliable estimate of their accuracy (uncertainty quantification). Because models of complex biological systems are always large simplifications, model discrepancy arises - where a mathematical model fails to recapitulate the true data generating process. This presents a particular challenge for making accurate predictions, and especially for making accurate estimates of uncertainty in these predictions. Experimentalists and modellers must choose which experimental procedures (protocols) are used to produce data to train their models. We propose to characterise uncertainty owing to model discrepancy with an ensemble of parameter sets, each of which results from training to data from a different protocol. The variability in predictions from this ensemble provides an empirical estimate of predictive uncertainty owing to model discrepancy, even for unseen protocols. We use the example of electrophysiology experiments, which are used to investigate the kinetics of the hERG potassium ion channel. Here, 'information-rich' protocols allow mathematical models to be trained using numerous short experiments performed on the same cell. Typically, assuming independent observational errors and training a model to an individual experiment results in parameter estimates with very little dependence on observational noise. Moreover, parameter sets arising from the same model applied to different experiments often conflict - indicative of model discrepancy. Our methods will help select more suitable mathematical models of hERG for future studies, and will be widely applicable to a range of biological modelling problems.

The importance of numax (the frequency of maximum oscillation power) for asteroseismology has been demonstrated widely in the previous decade, especially for red giants. With the large amount of photometric data from CoRoT, Kepler and TESS, several automated algorithms to retrieve numax values have been introduced. Most of these algorithms correct the granulation background in the power spectrum by fitting a model and subtracting it before measuring numax. We have developed a method that does not require fitting to the granulation background. Instead, we simply divide the power spectrum by a function of the form nu^-2, to remove the slope due to granulation background, and then smooth to measure numax. This method is fast, simple and avoids degeneracies associated with fitting. The method is able to measure oscillations in 99.9% of previously-studied Kepler red giants, with a systematic offset of 1.5 % in numax values that that we are able to calibrate. On comparing the seismic radii from this work with Gaia, we see similar trends to those observed in previous studies. Additionally, our values of width of the power envelope can clearly identify the dipole mode suppressed stars as a distinct population, hence as a way to detect them. We also applied our method to stars with low (0.19--18.35 muHz) and found it works well to correctly identify the oscillations.

In recent years, predicting mobile app usage has become increasingly important for areas like app recommendation, user behaviour analysis, and mobile resource management. Existing models, however, struggle with the heterogeneous nature of contextual data and the user cold start problem. This study introduces a novel prediction model, Mobile App Prediction Leveraging Large Language Model Embeddings (MAPLE), which employs Large Language Models (LLMs) and installed app similarity to overcome these challenges. MAPLE utilises the power of LLMs to process contextual data and discern intricate relationships within it effectively. Additionally, we explore the use of installed app similarity to address the cold start problem, facilitating the modelling of user preferences and habits, even for new users with limited historical data. In essence, our research presents MAPLE as a novel, potent, and practical approach to app usage prediction, making significant strides in resolving issues faced by existing models. MAPLE stands out as a comprehensive and effective solution, setting a new benchmark for more precise and personalised app usage predictions. In tests on two real-world datasets, MAPLE surpasses contemporary models in both standard and cold start scenarios. These outcomes validate MAPLE's capacity for precise app usage predictions and its resilience against the cold start problem. This enhanced performance stems from the model's proficiency in capturing complex temporal patterns and leveraging contextual information. As a result, MAPLE can potentially improve personalised mobile app usage predictions and user experiences markedly.

Learning PDE dynamics with neural solvers can significantly improve wall-clock efficiency and accuracy compared with classical numerical solvers. In recent years, foundation models for PDEs have largely adopted multi-scale windowed self-attention, with the scOT backbone in \textsc{Poseidon} serving as a representative example. However, because of their locality, truly globally consistent spectral coupling can only be propagated gradually through deep stacking and window shifting. This weakens global coupling and leads to error accumulation and drift during closed-loop rollouts. To address this, we propose \textbf{DRIFT-Net}. It employs a dual-branch design comprising a spectral branch and an image branch. The spectral branch is responsible for capturing global, large-scale low-frequency information, whereas the image branch focuses on local details and nonstationary structures. Specifically, we first perform controlled, lightweight mixing within the low-frequency range. Then we fuse the spectral and image paths at each layer via bandwise weighting, which avoids the width inflation and training instability caused by naive concatenation. The fused result is transformed back into the spatial domain and added to the image branch, thereby preserving both global structure and high-frequency details across scales. Compared with strong attention-based baselines, DRIFT-Net achieves lower error and higher throughput with fewer parameters under identical training settings and budget. On Navier--Stokes benchmarks, the relative $L_{1}$ error is reduced by 7\%--54\%, the parameter count decreases by about 15\%, and the throughput remains higher than scOT. Ablation studies and theoretical analyses further demonstrate the stability and effectiveness of this design. The code is available at this https URL.

Contrastive Language-Image Pretraining (CLIP) excels at learning generalizable image representations but often falls short in zero-shot inference on certain downstream datasets. Test-time adaptation (TTA) mitigates this issue by adjusting components like normalization layers or context prompts, yet it typically requires large batch sizes and extensive augmentations, leading to high computational costs. This raises a key question: Can VLMs' performance drop in specific test cases be mitigated through efficient, training-free approaches? To explore the solution, we investigate token condensation (TC) techniques, originally designed to enhance vision transformer efficiency by refining token usage during inference. We observe that informative tokens improve visual-text alignment in VLMs like CLIP on unseen datasets. However, existing TC methods often fail to maintain in-distribution performance when reducing tokens, prompting us to ask: How can we transform TC into an effective ``free-lunch'' adaptation strategy for VLMs? To address this, we propose Token Condensation as Adaptation (TCA), a training-free adaptation method that takes a step beyond standard TC. Rather than passively discarding tokens, TCA condenses token representation by introducing reservoir-based domain anchor tokens for information-preserving token reduction and logits correction. TCA achieves up to a 21.4% performance improvement over the strongest baseline on cross-dataset benchmark and the CIFAR-100-Corrupted dataset while reducing GFLOPs by 12.2% to 48.9%, with minimal hyperparameter dependency on both CLIP and SigLIP series.