Landscape feature representations play a central role in automated algorithm selection and meta-learning for black-box optimization, yet little is known about how different representations agree (or disagree) in the structures they impose on problem spaces. This paper presents a systematic unsupervised evaluation of four state-of-the-art representations (ELA, DeepELA, TransOptAS, and DoE2Vec) using a diverse set of affine combinations of BBOB functions (MA-BBOB). By applying extensive clustering analyses, coverage-based stability measures, and cross-representation similarity assessments, we show that each representation organizes the same problems in markedly different ways: ELA and TransOptAS form compact geometric structures, DeepELA provides a balanced intermediate view, and DoE2Vec achieves strong semantic alignment but with substantial fragmentation. Our results reveal that no single representation dominates; rather, they capture complementary aspects of the underlying landscapes. These findings highlight the importance of multi-view analyses for understanding representation behavior and offer guidance on selecting or combining representations in downstream meta-learning and algorithm selection tasks. In addition, across two different algorithm families (Differential Evolution and Particle Swarm Optimization), we show that landscape representations face an inherent trade-off in how well they align structural landscape descriptions with observed performance, indicating that no single representation can fully capture algorithm performance.

Algorithm Selection (AS) aims to automatically identify the most suitable optimization algorithm for a given problem instance by leveraging measurable problem characteristics and historical performance data. In this study, we investigate the generalization ability of AS models across both synthetic and real-world optimization landscapes. We consider two widely used academic benchmark suites (BBOB and CEC) and two real-world problem sets (robotics trajectory optimization tasks and unmanned aerial vehicle path-planning problems). Through a systematic cross-benchmark evaluation, we analyze how AS models transfer between domains, identify where generalization succeeds or breaks down, and highlight the challenges that arise when applying AS in realistic, domain-specific contexts. Our findings provide insights into the robustness of current AS approaches and inform the development of more reliable, broadly applicable AS systems for real-world optimization.

Automated algorithm selection in black-box optimization typically relies on supervised models that map landscape features to algorithm performance labels. Such models are costly to train, benchmark-dependent, and often fail to generalize to unseen problem classes. We study an unsupervised alternative: multi-kernel clustering over heterogeneous landscape representations, in which problem instances are grouped without using performance labels in the clustering stage, and the resulting clusters are mapped post hoc to solver recommendations through a strictly separated three-stage evaluation protocol. Drawing on two decades of advances in multiple kernel learning, we adopt a multi-kernel k-means formulation that jointly learns cluster assignments and kernel weights over four heterogeneous landscape views: ELA, DeepELA, DoE2Vec, and TransOptAS. On affine BBOB-derived selector tasks for Differential Evolution (DE) and Particle Swarm Optimization (PSO) at a fixed evaluation budget, we report mean±standard-deviation selector profiles over 50 independent random seeds for stochastic configurations. Multi-kernel clustering obtains the strongest mean profile on the DE portfolio and remains competitive with, and nominally ahead of, the leading baselines on the more compressed PSO portfolio, where differences among the best methods are small relative to stochastic variation. In representative median-seed runs used for visualization, the learned kernel weights retain ELA and TransOptAS while assigning zero weight to DeepELA and DoE2Vec, providing a task-specific interpretation of which representations are retained by the multi-kernel model for selector-oriented grouping.

While reusing treated wastewater (TWW) for irrigation provides a sustainable solution to water scarcity, it can potentially introduce contaminants that may threaten crop safety and quality. Consequently, further research is needed to understand the effects of TWW on the metabolomic and elemental profiles of irrigated crops. This study investigated how using TWW affect metabolism and element uptake in tomatoes grown in soil (lysimeters) and soilless (hydroponics) systems. Soil-grown tomatoes were irrigated with potable water, treated wastewater, and treated wastewater spiked with 14 CECs (0.1 mg/L), which included bisphenols, non-steroidal anti-inflammatory drugs, estrogens, and caffeine. Hydroponically grown tomatoes were grown in a nutrient solution, with or without CECs. Tomatoes were assessed by analysing sugars, organic acids, polyphenols, carotenoids, amino acids, fatty acids, and elements. Classification machine learning models were applied, and the best-performing model, a decision tree classifier, achieved 88% accuracy in distinguishing treatments under stratified five-fold cross-validation. The SHAP method identified key metabolites (ascorbic, palmitic, margaric, oleic, linoleic, behenic acids) and elements (Cd, Co, Cs, Cu, P, Na) that drive treatment differentiation. This study demonstrates how explainable machine learning can decode complex metabolic interactions, providing insights into the effects of using treated wastewater in agriculture.

Background:Reliable nutrient profiling and semantic interoperability are essential for scalable dietary assessment, food labeling (e.g., traffic-light schemes), and FAIR integration of food composition and consumption data. However, general-purpose large language models (LLMs) are not systematically exposed to structured recipe-nutrition mappings and food ontologies, limiting their accuracy and trustworthiness in food and nutrition tasks. Scope and approach:We review recent LLM advances in life sciences and healthcare and analyze the gap in food and nutrition applications. To address this gap, we introduce FoodyLLM, a domain-specialized LLM fine-tuned on 225k task-aligned QA pairs for (i) recipe nutrient estimation, (ii) traffic-light classification, and (iii) ontology-based entity linking to support FAIR food data interoperability. We benchmark FoodyLLM against strong general-purpose baselines (e.g., Llama 3 8B, Gemini 2.0) under zero-/few-shot prompting across five evaluation folds. Key findings:Across all tasks, FoodyLLM substantially outperforms general-purpose LLMs for nutrient estimation across all macronutrients (fat, protein, salt, saturates, sugar), accuracy increases from 0.43 to 0.63 to 0.91-0.97; for traffic-light classification across all nutrients and color categories, macro F1 improves from 0.46 to 0.80 to 0.86-0.97; and for ontology-based food entity linking across FoodOn, SNOMED-CT, and Hansard, macro F1 increases from 0.33 to 0.44 (best general-purpose baseline) to 0.93-0.98 on artificial NEL data, and from 0.24 to 0.51 to 0.67-0.84 on real corpora (CafeteriaSA and CafeteriaFCD). Overall, our results demonstrate the practical value of domain-specialized LLMs in food and nutrition research. They enable automated dietary assessment, large-scale nutritional monitoring, and FAIR data integration, while opening new pathways toward sustainable and personalized nutrition.

Biomedical and food entity normalization maps textual mentions to stable ontology identifiers, enabling literature mining and clinical data integration. However, continuous ontology updates degrade supervised models tightly coupled to specific vocabulary versions. We propose OntoRAG, a training- free normalization methodology that decouples semantic reasoning from the target knowledge base. Our methodology consists of several modules: hybrid retrieval for candidate generation, rank fusion for consolidating retrieval evidence, large language model (LLM) selection, confidence-gated stopping, and iterative query reformulation. We evaluate the methodology across four benchmark domains: CRAFT ChEBI (chemistry), NCBI Disease (medicine), NLM-Gene (genomics), and CafeteriaFCD (food and nutrition). Across evaluations of several state-of-the-art LLM models (a single one used for all modules in the methodology), the strongest one achieves exact-match accuracies of 84.3% (CRAFT ChEBI), 76.3% (CafeteriaFCD), 74.3% (NCBI Disease), and 64.7% (NLM-Gene). A component-specialized mixed-model configuration improves macro accuracy from 74.9% to 77.3%, yielding a substantial gain of 2.3 percentage points and a 9.6% reduction in classification error. Performance decomposition reveals that end-to-end accuracy is fundamentally bounded by candidate generation limits rather than contextual reasoninglimits. Controlled ablations confirm that lexical retrieval and iterative query reformulation drive overall accuracy. By treating the target ontology as an interchangeable offline index, OntoRAG provides a zero-shot normalization layer that adapts instantly to new vocabulary releases without task-specific retraining.

Benchmarking deep learning (DL) models for multi-label classification (MLC) of remote sensing images (RSI) typically yields rankings that do not generalize beyond the evaluated datasets. In this work, we move beyond rankings by employing functional analysis of variance (fANOVA) to systematically quantify the contributions of individual design choices and their interactions to performance variability. We conduct two empirical analyses covering 48 and 20 DL models, respectively, spanning design choices such as network architecture, fine-tuning strategy, learning strategy, and initialization. By applying fANOVA across seven MLC RSI datasets, we construct dataset meta-representations that capture design-choice sensitivity profiles. Hierarchical clustering of these meta-representations reveals that datasets naturally group according to how they respond to design decisions, with patterns strongly linked to intrinsic dataset properties such as scale, spatial resolution, and label space complexity. Our findings show that for large-scale datasets, fine-tuning strategy and architecture are dominant factors, while in data-limited regimes, initialization becomes decisive. For intermediate regimes, the interaction between architecture and learning strategy governs performance.

General-purpose Large Language Models (LLMs) like Llama, GPT, and Mistral struggle with domain-specific challenges in food and nutrition, where data is fragmented, heterogeneous, and semantically complex. While fine-tuned LLMs have shown success in healthcare and life sciences, similar progress in food domains has been limited, largely due to the lack of high-quality, task-specific datasets. We present FoodBench, a curated benchmark dataset of question–answer pairs designed for training and evaluating LLMs in food and nutrition. It spans key tasks such as nutrient estimation, food traffic-light classification, synonym linking, cooking measurement conversion, and food named-entity recognition and linking. FoodBench enables robust performance evaluation across zero-, one-, and few-shot settings, laying the groundwork for trustworthy, domain-adapted language models. This resource supports advances in personalized nutrition, dietary assessment, and food system innovation. Evaluation of four general-purpose LLMs (Llama 3, Mistral, Gemma, Gemini) on FoodBench tasks shows limited performance across nutrient estimation, traffic-light classification, and food interoperability, even with few-shot prompting. These results highlight the need for domain-specialized LLMs fine-tuned on food data, while establishing FoodBench as a benchmark not only for assessing general-purpose models but also for guiding and evaluating fine-tuning efforts.

Benchmarking shortest-path algorithms is commonly based on aggregate performance over heterogeneous graph sets, which limits insight into how different search paradigms react to instance structure. We adopt an instance-landscape view of graph benchmarking by embedding graphs into a low-cost structural feature space and clustering them into regions of similar structure. Three benchmark suites are studied: weighted Erdős–Rényi graphs, random geometric (wireless) graphs, and real-world road networks. We evaluate four representative shortest-path solvers spanning uninformed exact search (Dijkstra), bidirectional exact search (bidirectional Dijkstra), heuristic-guided exact search (A^*), and deque-based strategies (DEQ). Clustering robustness is analyzed under multiple feature-selection schemes, and runtime distributions are compared across landscape regions using non-parametric tests. While generator parameters induce stable structural regions, we find that feature-space similarity does not necessarily imply performance similarity: significant runtime shifts are frequently observed even within the same landscape region. A merged-suite analysis further shows that different benchmark families occupy largely disjoint regions. These results highlight both the potential and the limits of structural landscapes for the structure-aware benchmarking of shortest-path algorithms.

Multilingual text embedding models enable cross-lingual transfer of knowledge across a wide range of NLP tasks, but their evaluation remains highly uneven across high-, mid- and low-resource languages. In this paper, we propose a two-dimensional framework, specifically tailored for analyzing multilingual embedding benchmarks under dataset scarcity, and apply it on the Slavic-language subset of the MTEB benchmark. The framework distinguishes between task-specific and cross-task evaluation, while jointly analyzing three complementary aspects: (1) ranking robustness, (2) model consistency, and (3) evidence strength. At the task-specific level, we evaluate the stability of model rankings under changes in ranking methodology and benchmark dataset composition. At the cross-task level, we assess the ability of models to generalize across diverse tasks within a language. To quantify the reliability of benchmark conclusions, we introduce an Evidence Strength Score that accounts for dataset availability, diversity, and robustness assessability. Our analysis reveals severe benchmark sparsity, with many Slavic language-task pairs relying on a single dataset or highly correlated benchmark collections, limiting the ability to draw robust conclusions. The cross-task analysis reveals a small group of highly transferable models, most notably llama-embed-nemotron-8b, multilingual-e5-large-instruct, and Qwen3-Embedding variants, that consistently perform well across Slavic languages and tasks. Overall, the results demonstrate that benchmark rankings and robustness conclusions must be interpreted jointly with certain notation of their evidence strength and highlight benchmark scarcity as a major obstacle to trustworthy multilingual evaluation.

In large-scale benchmarking of stochastic optimization algorithms, the key challenge is no longer whether repeated runs are needed for reliability, but how to determine when sufficient evidence has been collected without incurring unnecessary computational cost. We study a learning-based extension of a recent empirical online heuristic that adaptively estimates the required number of runs using outlier handling and skewness-based symmetry checks. Using annotated outcomes from 132,000 Nevergrad runs on COCO (24 problems in 20 dimensions, 10 instances each, 11 optimizers), we train classifiers on 23 statistical, energy-free, and shape and stability features to predict whether a run-number estimate is reliable, prioritizing detection of incorrect estimates via minority-class recall. We evaluate reliability prediction using a within-configuration learning setup, where models are trained and tested on data sharing the same optimizer. The results show that run-number reliability can be learned in a within-configuration scenario, enabling detection of unreliable estimates with high minority-class recall, although performance remains limited by the restricted data diversity within fixed configurations.

This survey examines key advancements in designing features to represent optimization problem instances, algorithm instances, and their interactions within the context of single-objective continuous black-box optimization. These features support machine learning tasks such as algorithm selection, algorithm configuration, and problem classification, and they are also used to evaluate the complementarity of benchmark problem sets. We provide a comprehensive overview of problem landscape features, algorithm features, high-level problem-algorithm interaction features, and trajectory features, including the latest works from the past five years. We also point out limitations of the current state-of-the-art and suggest directions for future research.

The representation of optimization problems and algorithms in terms of numerical features is a well-established tool for comparing optimization problem instances, for analyzing the behavior of optimization algorithms, and the quality of existing problem benchmarks, as well as for automated per-instance algorithm selection and configuration approaches. Extending purely problem-centered feature collections, our recently proposed DynamoRep features provide a simple and inexpensive representation of the algorithmproblem interaction during the optimization process. In this paper, we conduct a comprehensive analysis of the predictive power of the DynamoRep features for the problem classification, algorithm selection, and algorithm classification tasks. In particular, the features are evaluated for the classification of problem instances into problem classes from the BBOB (Black Box Optimization Benchmarking) suite, selecting the best algorithm to solve a given problem from a portfolio of three algorithms (Differential Evolution, Evolutionary Strategy, and Particle Swarm Optimization), as well as distinguishing these algorithms based on their trajectories. We show that, despite being much cheaper to compute, they can yield results comparable to those using state-ofthe-art Exploratory Landscape Analysis features.

Large-scale multilingual text embedding models play crucial role in both research and industry, yet their behavior in language-specific, multi-task settings remains insufficiently understood. Although benchmarking platforms such as MTEB report results across more than 250 languages, conclusions about model superiority often depend on implicit choices of dataset compositions and performance aggregation methods. To address this gap, we present a meta-study of multilingual model performance robustness in MTEB, applying a diverse set of multi-criteria decision-making ranking schemes and introducing two robustness indicators: dataset-composition robustness (sensitivity of rankings to changing dataset compositions) and ranking-scheme robustness (sensitivity to aggregation method change). They enable systematic sensitivity analysis of whether benchmarking conclusions remain stable under different evaluation designs. We conduct an in-depth analysis on five languages (English, French, German, Hindi, and Spanish) across nine tasks (e.g., classification, clustering, retrieval) and release results for approximately 230 additional languages. The task-specific analyses show that large-scale LLM-based models are often robust top performers, though not uniformly (e.g., in retrieval task), while task-agnostic results reveal that only a small subset of models remains consistently strong across tasks, ranking schemes, and data subsamples.

The PSO-X framework incorporates dozens of modules that have been proposed for solving single-objective continuous optimization problems using particle swarm optimization. While modular frameworks enable users to automatically generate and configure algorithms tailored to specific optimization problems, the complexity of this process increases with the number of modules in the framework and the degrees of freedom defined for their interaction. Understanding how modules affect the performance of algorithms for different problems is critical to making the process of finding effective implementations more efficient and identifying promising areas for further investigation. Despite their practical applications and scientific relevance, there is a lack of empirical studies investigating which modules matter most in modular optimization frameworks and how they interact. In this paper, we analyze the performance of 1424 particle swarm optimization algorithms instantiated from the PSO-X framework on the 25 functions in the CEC'05 benchmark suite with 10 and 30 dimensions. We use functional ANOVA to quantify the impact of modules and their combinations on performance in different problem classes. In practice, this allows us to identify which modules have greater influence on PSO-X performance depending on problem features such as multimodality, mathematical transformations and varying dimensionality. We then perform a cluster analysis to identify groups of problem classes that share similar module effect patterns. Our results show low variability in the importance of modules in all problem classes, suggesting that particle swarm optimization performance is driven by a few influential modules.

Evaluating large language models (LLMs) across comprehensive benchmarks is expensive and time-consuming. We propose a graph-based prompt selection framework that models each benchmark as a similarity graph – nodes are prompts connected if their embedding-space distance falls above a configurable threshold – and applies Maximum Independent Set (MIS) algorithms to select a maximally diverse, non-redundant subset. We evaluate four MIS solvers (CPLEX, GREEDY, Online-MIS, ReduMIS) across six embedding models, three distance measures, six percentile thresholds, and four benchmarks (GPQA, IFEval, MMLU-Pro, Omni-MATH) covering 66 LLMs. Our central hypothesis – that repeated selection under different random seeds yields consistent LLM rankings that may also differ from the full-benchmark baseline – is strongly confirmed: Kendall's W ≥ 0.90 in 99.2% of stochastic configurations (mean W = 0.997 ± 0.008), while at higher percentile thresholds selected subsets achieve 25–48% prompt reduction on average. Ranking divergence from the full benchmark (ρ< 0.95) occurs in only 15.95% of configurations, concentrated at low thresholds (p_10–p_20) and benchmarks (GPQA, IFEval), identifying overly dense graphs as the primary failure mode.

Modular algorithm frameworks not only allow for combinations never tested in manually selected algorithm portfolios, but they also provide a structured approach to assess which algorithmic ideas are crucial for the observed performance of algorithms. In this paper, we propose a methodology for analyzing the impact of the different modules on the overall performance. We consider modular frameworks for two widely used families of derivative-free, black-box optimization algorithms, the covariance matrix adaptation evolution strategy (CMA-ES) and differential evolution (DE). More specifically, we use performance data of 324 modCMA-ES and 576 modDE algorithm variants (with each variant corresponding to a specific configuration of modules) obtained on the 24 BBOB problems for six different runtime budgets in two dimensions. Our analysis of these data reveals that the impact of individual modules on overall algorithm performance varies significantly. Notably, among the examined modules, the elitism module in CMA-ES and the linear population size reduction module in DE exhibit the most significant impact on performance. Furthermore, our exploratory data analysis of problem landscape data suggests that the most relevant landscape features remain consistent regardless of the configuration of individual modules, but the influence that these features have on regression accuracy varies. In addition, we apply classifiers that exploit feature importance with respect to the trained models for performance prediction and performance data, to predict the modular configurations of CMA-ES and DE algorithm variants. The results show that the predicted configurations do not exhibit a statistically significant difference in performance compared to the true configurations, with the percentage varying depending on the setup (from 49.1% to 95.5% for modCMA and 21.7% to 77.1% for DE).

This study presents an in-depth analysis of module importance within the modular CMA-ES (modCMA-ES) algorithm using exploratory data analysis and large-scale benchmarking across the BBOB suite. Rather than introducing new algorithms, our contribution lies in uncovering how individual modules and their interactions influence optimization performance across diverse black-box problem classes. We evaluate 324 modCMA-ES variants across 24 problem classes using functional ANOVA (f-ANOVA) to quantify the variance in performance attributable to individual, pairwise, and triplet module interactions. Results reveal substantial variation in module importance across problem classes and highlight strong alignment between module interaction patterns and high-level landscape features, particularly multi-modality. Further, we demonstrate that configuring only the most important modules - identified via f-ANOVA - achieves performance comparable to or better than the single-best solver, especially in high-dimensional settings. This analysis, conducted at both low (5D) and high (30D) dimensions, offers actionable insights into module interactions within the mod-CMA-ES framework.