The scale of the undertaking is striking. Running the final query against Web of Science on 27 January 2023 yielded 83,950 entries; the equivalent Scopus search returned 98,574. After merging the two lists and removing duplicates based on DOIs, the refined database settled at 108,064 unique records — down from an original 182,323. Every entry carries twelve variables, including an internal identification number, article type, author information, title, publication year, journal details, DOI, abstract, author keywords, indexed keywords from Scopus, keywords generated by Web of Science, and information about duplicates and database origin. The whole dataset is stored in the Zenodo repository under the Open Data Commons Attribution License, meaning any research team, anywhere, can sub-query it for their own purposes.

Why build such a shared foundation? The motivation lies in the nature of the science itself. Ecosystem services — the benefits humans derive from nature, spanning provisioning, regulating and cultural categories — have become central to academic research and decision-making alike, particularly as climate change, land degradation and urbanisation intensify. Closely connected is the assessment of ecosystem condition, the evaluation of the quality, state or health of an ecosystem, which determines its capacity to yield services. The latest developments in both fields converge in ecosystem accounting, an integrated statistical framework provided by the System of Environmental-Economic Accounts (SEEA) Ecosystem Accounting, adopted by the United Nations, that measures ecosystem assets, their condition and the flows of services into human activity.

SELINA aims to provide robust information and guidance for stakeholder groups to support transformative change in the EU — halting biodiversity decline, supporting ecosystem restoration and securing the sustainable supply of essential ecosystem services by 2030. Yet recent literature reviews showed that the uptake of ecosystem service assessments in decision-making remains limited, both in documentation and in practice. One of the project’s motivations is therefore to improve assessment design and communication of results tailored to stakeholder needs. The project builds on the Mapping and Assessment of Ecosystems and their Services (MAES) initiative, which under the EU Biodiversity Strategy 2020 provided the conceptual, methodological, data and knowledge base for comprehensive assessments at EU-wide and Member State scales, and follows the earlier ESMERALDA EU Horizon 2020 project (2015–2018).

Within SELINA, three large systematic literature reviews are underway, each with a distinct objective: deriving a minimum set of key ecosystem condition indicators per ecosystem type; revealing how uncertainties in ecosystem service assessments relate to the likelihood of uptake in decision-making; and identifying indicators that integrate information on both ecosystem condition and services. Rather than letting each review team build its own search from scratch, the project’s task leads agreed early in 2023 to establish a common Super-Query — a broad yet precisely structured search strategy serving multiple systematic reviews simultaneously, ensuring all three tasks draw from a harmonised and comprehensive dataset.

The methodology followed a slightly modified PRISMA protocol, the standard for systematic reviews, in seven steps: creating a task force of eight members drawn from the review leads and the SELINA Knowledge and Data Flows team, with expertise spanning environmental science, landscape ecology, landscape planning, physical geography and environmental economics; defining the scope; identifying key concepts, keywords and synonyms; developing the query string; fine-tuning and testing; running and recording the final query; and processing the harvested literature into the final database. The task force agreed to apply the Super-Query to Web of Science and Scopus, restricted to English-language, peer-reviewed articles published between 2018 and 2022 — capturing the most recent developments since ESMERALDA concluded.

The technical construction of the query reveals the depth of the challenge. Three general concepts were addressed: ecosystem services and nature’s contributions to people; ecosystem condition; and ecosystem accounting. For each, the team compiled general keywords, synonyms, spelling considerations and truncations with wildcards to capture word variations — for example, adding various terrestrial and aquatic ecosystem types as alternative terms for ecosystem condition. Controlled vocabulary and restrictive terms were included to limit findings to genuine assessments rather than papers that merely drop the terminology. The restrictive terms for ecosystem services list assessment methods, adopted from the IPBES values assessment and extended by 36 methods identified in ESMERALDA; those for ecosystem condition relate to specific measurement and evaluation keywords.

The iterative refinement process was equally revealing. An online tabular spreadsheet with semi-automated functionality merged individual keyword sections through predefined formulae, dynamically updating query combinations whenever keyword sets were modified, and even providing estimates of the potential volume of literature items per combination. An automated translation mechanism converted Web of Science query syntax into the corresponding Scopus format, adjusting field tags and operators — for instance, converting the proximity operator NEAR/10 to Scopus’s W/10. Notably, the two databases define their combined search fields differently: Web of Science’s Topic Search field includes Author Keywords and KeyWords Plus, terms generated from the titles of cited references, while Scopus’s TITLE-ABS-KEY field includes Author Keywords, Index Terms, Trade Names and Chemical Names, leading to differences in the scope and nature of results retrieved from each.

Fine-tuning exposed a common pitfall of large literature searches: many papers use ecosystem service or ecosystem condition concepts in their keywords without properly addressing them in content. Initially, querying ecosystem condition terms alone yielded more than one million entries; adding restrictive terms aimed at ensuring a proper assessment reduced the number to one tenth. The final search string comprises six search units — ecosystem services and nature’s contributions to people, ecosystem service synonyms, ecosystem service restrictions, ecosystem accounting, ecosystem condition, and ecosystem condition restrictions — all connected with the Boolean operator AND, with terms within each unit generally connected by OR. In two units, the NEAR/10 proximity operator connects ecosystem types to analogies of condition or service, maintaining semantic accuracy across the vast keyword space.

The profound collaboration of review task leads in this early phase of the project led to the successful compilation of the operational Super-Query database, a fundamental resource for the succeeding literature reviews. Review task leads will continue to sub-query relevant literature and proceed with their specific objectives, while the agreement on consistent terminology lays an important foundation for further collaboration within the consortium. Approximately 6,200 entries without a DOI remain in the database, so users are advised to perform additional duplicate removal based on identical titles after sub-querying. Beyond SELINA, the co-creation process has created a valuable resource to be used by the project and beyond — a single, transparent, reproducible search strategy that any team studying ecosystem condition, services or accounting can now build upon.

Subject of Research: Development of the "SELINA Super-Query" in a pan-European Horizon Europe project: A systematic literature search on Ecosystem Condition, Ecosystem Services and Ecosystem Accounting

Article Title: Development of the "SELINA Super-Query" in a pan-European Horizon Europe project: A systematic literature search on Ecosystem Condition, Ecosystem Services and Ecosystem Accounting

Article References: Seguin, J., Lange, S., Barton, D., Grêt-Regamey, A., Rendón, P., Roche, P., Nicholson Thomas, I., & Burkhard, B. (2025). Development of the "SELINA Super-Query" in a pan-European Horizon Europe project: A systematic literature search on Ecosystem Condition, Ecosystem Services and Ecosystem Accounting. One Ecosystem, 10, Article e155968. https://doi.org/10.3897/oneeco.10.e155968

Image Credits: AI Generated

DOI: 10.3897/oneeco.10.e155968

Keywords: Development, SELINA, Super-Query, pan-European, Horizon, Europe, project, systematic, literature, search, Ecosystem, Condition

The MLCS problem sits at the heart of comparative genomics. When researchers align multiple DNA sequences to find shared motifs, trace evolutionary relationships, or annotate functional elements, they are implicitly hunting for common subsequences that survive across all the sequences being compared. The difficulty is computational: as the number of sequences grows, the search space of possible common subsequences explodes exponentially. Exact algorithms, which guarantee the optimal answer, become impractical beyond small instances, so the field has long relied on approximate methods that trade guaranteed optimality for speed. The catch, as the authors of the new study note, is that many existing approximate approaches process complex multi-sequence biological data without effectively incorporating contextual information from local sequence fragments, which can cause promising candidate paths to be overlooked and makes it hard to balance computational efficiency against solution quality.

The framework introduced by Huafeng Quan, Yimeng Li, Yiting Li, and Jie Hu of Guizhou University of Finance and Economics in Guiyang, China, addresses this weakness with three tightly coupled modules. The first is a lightweight Transformer encoder combined with contrastive learning that performs representation learning on local sequence fragments. Transformers, the architecture behind modern large language models, excel at capturing dependencies between elements in a sequence. Here, the encoder learns to embed short fragments of the input sequences into a numerical space where fragments that play similar roles in the overall alignment problem end up close together. Contrastive learning, a training strategy that pushes the model to distinguish between related and unrelated examples, sharpens these representations so that the downstream search can rely on them as meaningful signals rather than noisy features.

The second module governs how the search itself expands. Beam search, the classic heuristic at the core of the framework, explores a search space level by level, keeping only a fixed number of the most promising partial solutions at each step and discarding the rest. This pruning is what makes beam search fast, but it is also what makes it fragile: if the scoring rule used to rank candidates is crude, genuinely good paths can be pruned early and never recovered. The new framework introduces a non-dominated filtering strategy and a dual-criterion evaluation mechanism during candidate-path expansion. Non-dominated filtering borrows a concept from multi-objective optimization, where a solution is kept only if no other solution is at least as good on every criterion and strictly better on at least one. Applying this idea to the branching search means that candidates are screened along multiple dimensions simultaneously rather than collapsing everything into a single number too early.

The third module is a multidimensional scoring model that ranks candidate paths and guides the beam-search pruning. Two of the criteria named in the study are positional dispersion, abbreviated PD, which captures how the positions of a candidate subsequence are spread across the input sequences, and representation-based priority, or PR, which draws directly on the learned embeddings from the Transformer encoder. By combining these signals, the scoring model can favor paths that not only look locally strong but also sit in regions of the search space where the learned representations suggest further extension is likely to pay off. This is the conceptual bridge between the neural and heuristic halves of the framework: the network supplies context-aware judgments about fragments, and the search converts those judgments into concrete pruning decisions.

To evaluate the approach, the researchers ran experiments on randomly simulated DNA datasets and on real biological sequence fragments obtained from NCBI, the US National Center for Biotechnology Information. The real datasets were deliberately varied in character. One set consisted of bacteriophage sequences, viruses that infect bacteria, with relatively concentrated host backgrounds, meaning the sequences share a fair degree of evolutionary proximity. The other set comprised heterogeneous viral sequences drawn from different viral taxa and sources, a harder test in which shared subsequences are rarer and the alignment landscape is more fragmented. Testing across both concentrated and diverse sequence families is important because an algorithm that only works on closely related sequences has limited value for broad comparative studies.

The results, as reported in the study, show that under different sequence-length and sequence-number settings the proposed method generally obtained longer common subsequences than representative approximate algorithms while maintaining good computational efficiency. In practical terms, this means the framework found common threads running through the input sequences that rival methods missed, without paying an unacceptable price in running time. The authors attribute this to the combination of context-based sequential representation information with heuristic search strategies: the learned representations steer the search toward branches that a purely positional or greedy heuristic would deprioritize, while the beam-search machinery keeps the overall computation bounded.

Perhaps the most reassuring finding for practitioners is what happened on small-scale instances where exact solutions could still be computed. On those problems, the approximate solutions produced by the hybrid framework remained close to the exact optima. This kind of validation matters because approximate algorithms are easy to run but hard to trust; without checkpoints against ground truth, there is no way to know whether a longer subsequence reported by one method reflects genuine algorithmic advantage or simply luck on a particular dataset. By demonstrating near-optimal behavior where verification is possible, the study builds a case that its quality holds more broadly, even though exact verification becomes infeasible on the larger instances where the method is most needed.

The implications reach beyond a single benchmark problem. The MLCS formulation underlies tasks in sequence annotation, targeted resequencing, and the analysis of long non-coding RNAs, all areas flagged by the journal as related subjects for the paper. As next-generation sequencing continues to flood laboratories with data, the gap between what exact algorithms can handle and what real datasets demand keeps widening. Hybrid approaches that let learned models inform classical search, rather than replacing search entirely, offer a pragmatic middle path: they inherit the efficiency guarantees and interpretability of heuristic frameworks while gaining the pattern-recognition power of deep learning. The Guizhou team’s framework, supported by provincial research funding programs in China, is an open-access contribution to that growing trend, and its published version carries the permanent DOI 10.1186/s12859-026-06704-9 for researchers who wish to examine the methods and reproduce the experiments in full.

Subject of Research: Computational Biology

Article Title: A hybrid beam search framework with deep representation learning for the multiple longest common subsequence problem

Article References: Quan, H., Li, Y., Li, Y., & Hu, J. (2026). A hybrid beam search framework with deep representation learning for the multiple longest common subsequence problem. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06704-9

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06704-9

Keywords: hybrid, beam, search, framework, deep, representation, learning, multiple, longest, common, subsequence, problem

The research team, led by Sarah L. H. Kamhout of the Department of Psychology at Brigham Young University, recruited 80 adolescents between the ages of 14 and 18 from the general community. Participants completed detailed 24-hour dietary recalls using the National Cancer Institute’s Automated Self-Administered 24-hour dietary assessment tool, known as ASA-24, on two to three randomly assigned days that included both weekdays and weekends. The tool captured intake of 26 different micronutrients, ranging from familiar players like calcium, iron, and vitamin D to less-studied carotenoids such as lycopene, lutein, and zeaxanthin. Participants were notified of their assigned recall day only on the day itself, a design choice intended to prevent them from altering their eating habits in response to being observed.

Sleep was measured with unusual rigor for a study of this kind. Each participant wore a wrist-worn actigraph for 7 to 20 days, providing objective data on sleep onset time, sleep midpoint, total sleep duration, sleep efficiency, and wake after sleep onset, a measure known as WASO. Because adolescents are known to be unreliable reporters of their own sleep habits, actigraphy readings were cross-verified with daily sleep diaries. The Sadeh algorithm, a validated method for converting movement data into sleep-wake states, was used to score each night. Sleep efficiency was calculated as the ratio of time asleep to time in bed, and all time-based variables were converted to minutes after midnight to avoid circular calculations.

Beyond nightly sleep, the study also probed more trait-like dimensions of sleep health. Participants completed the Morningness-Eveningness Questionnaire to assess chronotype, their innate preference for early or late sleep and wake times, and the Epworth Sleepiness Scale for Children and Adolescents to gauge daytime sleepiness. Most notably, each subject underwent a six-hour dim light melatonin onset protocol, or DLMO, in the laboratory. After a three-to-five-day at-home sleep stabilization period, saliva samples were collected every 30 to 45 minutes under light levels below 5 lux, beginning five hours before and ending one hour after each participant’s preferred bedtime. DLMO was defined as the moment salivary melatonin concentrations exceeded 4 picograms per milliliter, a threshold previously validated in adolescents. The gap between DLMO and average sleep onset, called the phase angle, served as a measure of circadian alignment, with values between two and four hours considered clinically aligned.

The statistical approach was equally careful. Because dietary reference intakes differ between males and females due to distinct biological and metabolic needs, all analyses were performed separately by sex. The researchers used mixed effects models with random by-participant intercepts to account for multiple dietary intakes per participant, then applied backwards step-wise model fitting to remove micronutrients that did not improve overall model fit. Variance inflation factors below five confirmed low collinearity in all final models, and p-values were corrected for multiple comparisons using the Benjamini-Hochberg false discovery rate procedure, applied separately for daily and averaged models.

Against the researchers’ initial hypotheses, no single micronutrient emerged as significantly associated with improved sleep health across all models. Yet two nutrients showed the most consistent positive signals. Potassium, with coefficients ranging from -0.03 to 0.004 and p-values below 0.05, was linked to increased same-night sleep efficiency in both males and females, and to earlier sleep onset in females. Retinol, a specific chemical form of vitamin A, showed associations with better sleep efficiency in females and earlier sleep midpoint and onset in males, with coefficients ranging from -0.20 to 0.01 and p-values below 0.05. These findings echo a classic double-blind, placebo-controlled crossover study in nine adult males on potassium-deficient diets, where supplementation improved sleep efficiency and consolidation, and more recent work linking lower urinary potassium excretion to greater daytime sleepiness in adults.

On the other side of the ledger, zinc and vitamin B12 were the micronutrients most consistently linked with negative sleep outcomes. Zinc, with coefficients ranging from -0.54 to 3.51 and p-values below 0.10, was associated with later sleep onset in males on the same night, reduced average sleep efficiency in males, and increased daytime sleepiness in both sexes. Vitamin B12, with coefficients ranging from -0.96 to 7.22 and p-values below 0.05, was linked to later sleep midpoint in males, reduced sleep efficiency in females, and increased wake after sleep onset in males. Notably, mean vitamin B12 intake in the sample surpassed the recommended daily allowance roughly three-fold for males and five-fold for females, a pattern consistent with prior research suggesting vitamin B12 can have an alerting effect that delays sleep onset and shortens sleep duration.

The dietary data itself revealed a troubling picture of adolescent nutrition. Less than 2 percent of youth ages 14 to 18 meet recommended vegetable intake guidelines according to national survey data, and only 2.9 percent of children and adolescents meet daily whole grain requirements. In this sample, average intake of sodium, copper, selenium, vitamin C, niacin, vitamin B6, vitamin K, and vitamin B12 exceeded recommended daily intake ranges for both sexes, while average intake of calcium, magnesium, potassium, vitamin D, and vitamin A fell below recommended ranges. Over half of participants did not meet recommended vitamin D intake, a deficiency consistently associated with decreased sleep duration and poorer sleep efficiency in children and adolescents. Iron, frequently tied to sleep disorders such as restless legs syndrome in prior research, showed no significant associations in this study, possibly because the female sample was on average below the recommended daily allowance for iron, making it difficult to detect effects of adequate or excess consumption.

The researchers emphasize that their findings are exploratory signals rather than established causal relationships. The cross-sectional design limits causal inference, though the day-to-night matching of dietary intake with same-night sleep helps establish temporal precedence in some analyses. Self-reported dietary recalls introduce potential recall and desirability biases, and the ASA-24, while demonstrating strong agreement with interviewer-administered recalls, is not a perfect measure of true intake. Absorption and metabolism of micronutrients also vary by individual, influenced by storage levels, interactions between nutrients, and whether intake comes from food or supplements. Carotenoids, for instance, are absorbed less efficiently than retinol, and people with low stores of vitamins A, calcium, and iron increase their ability to absorb more of these nutrients, meaning individual adolescents may have exhibited different patterns of absorption related to their specific consumption patterns.

Despite these limitations, the study represents the first broad examination of micronutrient intake and sleep health in adolescents, a population where nearly 70 percent fail to obtain the recommended 8 to 10 hours of sleep and over 40 percent report sleeping less than 7 hours per night. The findings suggest that sleep may be sensitive to day-to-day shifts in potassium intake, that different chemical forms of vitamin A may play distinct roles in sleep timing, and that zinc and vitamin B12 warrant closer scrutiny for their potential negative associations with sleep during adolescence. The researchers call for experimental and longitudinal designs to clarify causal pathways and for interventions tailored to individual intake patterns, timing of consumption, and interactions between multiple nutrients. No universal, one-size-fits-all dietary solution for improving adolescent sleep emerged from this work, but the study opens a promising avenue for understanding how the food on a teenager’s plate might ripple through the architecture of their night.

Subject of Research: Nutrition and Sleep Science

Article Title: In search of vitamin sleep: cross-sectional analysis of vitamin and mineral intake and sleep health outcomes in adolescents

Article References: Kamhout, S. L. H., Bellini, S. G., Luke, S. G., Homer, K. J., Broadbent, M., & Duraccio, K. M. (2026). In search of vitamin sleep: cross-sectional analysis of vitamin and mineral intake and sleep health outcomes in adolescents. Frontiers in Nutrition, 13, Article 1978612. https://doi.org/10.3389/fnut.2026.1978612

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1978612

Keywords: adolescent sleep, micronutrients, potassium, retinol, zinc, vitamin B12, circadian rhythm, DLMO, actigraphy, sleep health, search, vitamin

The work, led by Oorappan G M of KGiSL Institute of Technology in Coimbatore and Lakshmanan S A of Amrita Vishwa Vidyapeetham in Chennai, introduces what the authors call an SSS–DDQN solution-space best-first search controller. The name is a mouthful, but the idea is elegant. At its heart lies a double deep Q-network, or DDQN, a reinforcement-learning architecture that learns to estimate the long-term value of corrective actions while deliberately reducing the overestimation bias that plagues standard Q-learning. Wrapped around that learned value function is SSS, a best-first search algorithm descended from the minimax tradition of adversarial game playing. Rather than diving depth-first through a tree of possible futures the way Alpha–Beta search does, SSS* maintains a priority queue of partial solutions and always refines the most promising cluster first. The authors argue that this shift from depth-first to best-first refinement is exactly what a robot needs when it cannot see its own state clearly and must choose corrective actions under sensor noise and dropout.

To understand why this matters, consider what the robot is actually up against. The researchers model a four-Mecanum-wheel robot as a planar rigid body whose posture is described by inertial-frame position, heading angle, and body-frame velocities. The dynamics include an inertia matrix, a nonlinear velocity-coupling term, and a damping matrix, all of which the controller must respect while distributing a generalized force to four independently driven wheels through the Mecanum force-mapping mechanism. The reference trajectory is a lemniscate, a figure-eight-like curve chosen because it continuously changes curvature and demands combined forward, sideways, and rotational motion, making it a rigorous test of omnidirectional tracking. On top of this, the simulation injects zero-mean sensor noise into position, heading, velocity, and yaw-rate measurements, and adds a 3 percent probability of position dropout, mimicking the temporary localization loss that real robots suffer from wheel slip, occlusion, or communication delay.

Because the true state is hidden, the framework begins with an extended Kalman filter, or EKF, that estimates the hidden robot state and maintains a belief mean and covariance. The EKF prediction step uses the nonlinear Mecanum robot model to forecast the next belief, while the correction step uses the innovation, the difference between the predicted and measured observation, and the Kalman gain to update the belief through a Joseph covariance update that preserves numerical stability. This belief state, rather than any raw measurement, is what the learning and search components operate on. The authors are careful to distinguish this physics-informed belief-consistent approach from latent-dynamics methods like MuZero or TD-MPC, which learn internal state representations through self-supervised consistency losses. Here, consistency is enforced by the Kalman filtering machinery itself, grounded in the nonlinear robot model and the measurement residual, rather than by an auxiliary loss function.

The DDQN component is trained with a 15-dimensional state vector that includes tracking errors, body-frame velocities, error rates, and reference trajectory information, and a discrete action space of 27 corrective actions spanning longitudinal force, lateral force, and yaw moment. The network architecture uses three fully connected hidden layers with 192, 192, and 128 neurons and ReLU activations, outputting 27 Q-values. Training employs prioritized experience replay with a priority exponent of 0.6, a discount factor of 0.985, and an exploration rate that decays from 1.0 to 0.05. The learned corrective action is added to a nominal model-based torque command, so the reinforcement-learning agent refines rather than replaces the physical controller. The reward function penalizes position error, heading error, velocity error, control effort, and actuator saturation, with a bonus for accurate tracking and a failure penalty for large deviations.

Where the proposed method diverges from its baselines is in how it uses the DDQN at search time. The SSS–DDQN controller ranks all 27 candidate actions using QMDP belief sampling, which averages DDQN Q-values over 15 belief samples drawn from the EKF covariance, and passes only the top-ranked actions to the SSS search. The search itself uses a three-ply MAX–MIN–MAX structure, where MAX nodes represent robot action selection and MIN nodes represent worst-case disturbance response. Each node in the OPEN priority queue carries a LIVE or SOLVED label, a bound value, a root action, accumulated cost, and a cluster identity. When a LIVE node reaches the search horizon, it is converted into a SOLVED node by blending the averaged DDQN value with a negative model-based tracking cost and adding a belief-uncertainty penalty proportional to the covariance trace. Beta-bound pruning prevents over-optimistic DDQN leaf estimates from dominating the final decision, delayed MIN-sibling refinement pushes less critical disturbance branches to lower priority, and a secondary lookahead checks for hidden finite-horizon disturbance effects before the final action is committed.

The authors compare this controller against four baselines under identical conditions: EKF-belief PER-DDQN, which uses filtered belief states directly; QMDP-PER-DDQN, which averages Q-values over belief samples; Alpha–Beta Minimax, which performs depth-first adversarial search; and AB–DDQN Hybrid, which combines Alpha–Beta search with DDQN leaf evaluation. The proposed SSS*–DDQN achieved the best overall ranking with a root-mean-square position error of 0.01906 meters, a mean absolute error of 0.016074 meters, and a cumulative reward of 1843.6. Compared with the AB–DDQN Hybrid baseline, it reduced the RMS position error by 12.93 percent and the mean absolute error by 17.80 percent. Statistical validation over multiple independent stochastic trials, each with different random seeds for noise, dropout, and disturbance, showed the lowest mean RMS error of 0.01847 plus or minus 0.00132 meters and the lowest mean absolute error of 0.01554 plus or minus 0.00128 meters, with narrow 95 percent confidence intervals. The EKF covariance trace converged to nearly identical values across all controllers, confirming that the performance differences arise from action-selection and search strategies rather than estimation uncertainty.

The search-efficiency numbers reveal how the controller achieves this. The proposed method refined 264,423 nodes, generated 264,423 additional nodes, and pruned 64.27 percent of them, far exceeding the 13.44 percent pruning ratio of Alpha–Beta Minimax and the 16.06 percent of AB–DDQN Hybrid. It also performed 36,383 secondary lookahead operations, a mechanism no baseline employs. An ablation study confirms that no single module alone drives the improvement: the EKF provides stable belief estimation, QMDP improves uncertainty-aware action evaluation, DDQN supplies learned value guidance, SSS* refines promising action clusters, beta-bound pruning reduces unnecessary expansion, and secondary lookahead improves final action robustness. The authors also verified with a fourth-order Runge–Kutta sensitivity check that the comparative conclusions are not artifacts of Euler integration, since all controllers were evaluated under identical numerical and sensing conditions.

Yet the study is candid about its costs. Single-step runtime profiling in a MATLAB–NVIDIA Isaac Sim co-simulation through a ROS 2 interface showed that the proposed controller required a mean runtime of 4,830.4 milliseconds per control step, with a 95th-percentile runtime of 6,587.9 milliseconds, against a 20-millisecond deadline for 50 Hz control. Component-level profiling revealed that SSS* OPEN-list refinement accounted for roughly 98.8 percent of total execution time, making it the dominant bottleneck rather than belief-state estimation. By contrast, EKF-belief PER-DDQN completed all steps within the deadline at a mean of 1.961 milliseconds. The authors therefore recommend adaptive search budgeting, uncertainty-triggered SSS* activation, reduced top-ranked action sets, fewer belief samples, parallel or GPU evaluation, and fallback to simpler controllers when computational resources are limited. The present full-search configuration is suitable for simulation-based validation, offline analysis, and supervisory decision-making, but not yet for deterministic real-time control on embedded hardware.

The broader significance lies in the demonstration that learned value functions and classical best-first search can be formally integrated at four levels: action ranking, node expansion, leaf-node evaluation, and final action selection. The DDQN does not replace the search process; it guides it by reducing the branching factor, evaluating terminal nodes, influencing pruning bounds, and supporting robust final action choice. This positions the work as a bridge between the AlphaGo lineage of value-guided tree search, which has transformed game playing, and the practical demands of mobile robot control under partial observability. The authors identify future work in hardware testing, real-time profiling, additional trajectory cases, and comparison with PPO, SAC, TD3, model-predictive control, and MCTS-based methods. If the computational bottleneck can be tamed, the approach could offer a template for robots that reason about their own uncertainty the way a chess engine reasons about its position, always refining the most promising line first.

Subject of Research: SSS–DDQN solution-space best-first search for robust trajectory tracking of Mecanum robots under partial observability

Article Title: SSS–DDQN solution-space best-first search for robust trajectory tracking of Mecanum robots under partial observability

Article References: G M, O., & S A, L. (2026). SSS–DDQN solution-space best-first search for robust trajectory tracking of Mecanum robots under partial observability. Frontiers in Robotics and AI, 13, Article 1896260. https://doi.org/10.3389/frobt.2026.1896260

Image Credits: AI Generated

DOI: 10.3389/frobt.2026.1896260

Keywords: DDQN, solution-space, best-first, search, robust, trajectory, tracking, Mecanum, robots, under, partial, observability

The feature space is the framework that turns each data sample into a vector a model can analyze, and the authors argue that the success of any machine learning pipeline depends heavily on how well important features are selected and generated. Feature generation, as a central technique, converts raw data into a refined feature space that facilitates model training and enhances its structure. Yet despite steady progress in automated feature selection and generation, two major challenges persist: explainability and cross-domain generalization. Methods that work brilliantly on one dataset often fail to transfer to another, and the features they produce can be opaque to the experts who must act on them.

One popular alternative has been to let deep learning handle feature engineering automatically rather than relying on hand-crafted techniques. Deep learning models have proven highly effective across a variety of fields, but they are often considered less effective when applied to tabular datasets, where sample sizes are modest, columns are heterogeneous, and the signal is spread across interactions that neural architectures capture inefficiently. The İzmir team therefore took a different route: instead of asking a giant network to internalize the data, they asked small language models to reach outside the dataset entirely and pull in knowledge from the web.

The methodology centers on retrieval-augmented generation, the technique that lets a language model consult external sources before answering. For each tabular dataset, the system formulates search queries derived from the column names, values and the classification task itself. A small language model—compact enough to run without massive infrastructure—then processes the retrieved web content and distills it into concrete, machine-readable features. These web-search-based features are appended to the original tabular columns, enriching each row with external context that no amount of resampling the raw data could have produced.

The choice of a small language model rather than a frontier-scale one is deliberate. Small models are cheaper to run, easier to deploy, and—crucially for the explainability problem—their prompts and retrieval steps can be inspected. The study also emphasizes prompt design as a first-class component of the pipeline: how the model is asked to read a web page and translate it into a feature determines whether the extracted signal is meaningful or noise. Because the resulting features are grounded in retrieved documents rather than in the latent whims of a neural network, a domain expert can, in principle, trace why a model believes a certain external fact is relevant.

Experimental results show the approach reaching highest F1 scores of 0.383 and 0.848 on two well-known tabular datasets for the machine learning multi-class classification task. The F1 score, the harmonic mean of precision and recall, is a demanding metric for multi-class problems, where a model must balance accuracy across every category rather than excelling on the majority class. The gap between the two scores reflects the reality of tabular classification: some datasets carry rich external context that web knowledge can illuminate, while others resist enrichment because their structure is idiosyncratic or their semantics are not well represented online.

The broader significance lies in what this says about the evolving role of language models in data science. Retrieval-augmented generation has mostly been framed as a question-answering technology—a way to make chatbots cite their sources. Inan and Oguz repurpose it as a feature factory. The web becomes a vast, continuously updated auxiliary dataset, and the small language model becomes the extraction engine that converts unstructured human knowledge into the structured columns that classical classifiers consume. This reframing matters because it suggests that the value of language models for tabular problems may lie not in replacing gradient-boosted trees, but in feeding them better inputs.

It also speaks to the open-source momentum in the field. The study’s keywords—feature selection, retrieval-augmented generation, small language models, open-source and prompt design—map onto a research agenda that many laboratories can pursue without access to proprietary giants. A team with a modest GPU and a careful prompt library can replicate the pipeline, adapt the search strategies to their domain, and audit every feature the system proposes. That transparency directly addresses the explainability gap the authors identify as a central challenge of automated feature engineering.

Cross-domain generalization remains the harder test. A feature generated from web content is only as good as the match between the dataset’s semantics and the public knowledge the search engine surfaces. Medical datasets may benefit from clinical literature, while proprietary industrial sensor logs may find little useful context online. The reported F1 scores, while promising on the stronger dataset, are an honest reminder that enriching tabular data with web knowledge is not a universal accelerant. The authors’ contribution is a methodology and a demonstration, not a guarantee—and they present it as an alternative to deep-learning-based automatic feature engineering rather than a replacement for domain expertise.

Still, the vision is arresting. Imagine a near future in which every spreadsheet arrives pre-enriched: a language model has already searched the web for what each column means, retrieved the relevant facts, and generated features that encode them. Classification models would train on data that carries not just measurements but context, and analysts would see, in plain retrieved documents, why each new column exists. Inan and Oguz have taken an early, concrete step toward that future, showing that small language models can select and generate web-search-based features for tabular classification—and that the humble spreadsheet may finally have found its companion in the age of language models.

Subject of Research: Feature selection and generation with small language models and web search for tabular data classification

Article Title: Selecting and generating web search based features using small language models for tabular data classification

Article References: Inan, E., & Oguz, D. (2026). Selecting and generating web search based features using small language models for tabular data classification. Scientific Reports. https://doi.org/10.1038/s41598-026-75169-w

Image Credits: AI Generated

DOI: 10.1038/s41598-026-75169-w

Keywords: Feature selection, Retrieval-augmented generation, Small language models, Tabular data, Machine learning, Open-source, Prompt design, Selecting, generating, search, based, features

The study, conducted by Marco Vasconcelos of the University of Aveiro, Cyrus Kirkman and Aaron P. Blaisdell of the University of California, Los Angeles, and Armando Machado of the University of Aveiro, is the first to systematically examine suboptimal choice in a sequential accept-reject framework. Rather than presenting two options side by side, the researchers asked a fundamentally different question: what happens when an animal encounters one option at a time and must decide whether to accept it or reject it in favor of an unseen alternative?

The experimental logic builds on the classic suboptimal choice paradigm, in which pigeons choose between two distinctially colored alternatives. One option, the informative or S+ alternative, leads to a stimulus that reliably signals whether food will or will not follow—providing perfect predictive information about the upcoming outcome. The other, the non-informative alternative, leads to stimuli that are equally likely to precede food regardless of the actual outcome. Although the informative option delivers food on only 50 percent of trials and the non-informative option delivers food on 75 percent of trials, pigeons in simultaneous choice tests consistently prefer the informative option, sacrificing overall reward for the psychological value of knowing what is coming.

What makes the new study remarkable is its methodological shift. In the sequential version of the task, pigeons first encountered one option—either the informative or the non-informative alternative—and were required to accept it, proceeding with that outcome, or reject it, which then delivered the alternative option instead. This structure transforms the task from a simple choice into a search problem, where each encountered option is evaluated in turn before a final decision is locked in. The researchers reasoned that if suboptimal preference is a genuine feature of the animal’s valuation system rather than an artifact of simultaneous presentation, it should also manifest in sequential contexts.

The results in the cost-free condition were striking. When rejecting an option carried no penalty other than delivering the alternative, pigeons overwhelmingly accepted the informative option whenever it appeared and just as overwhelmingly rejected the non-informative option, effectively routing themselves through the informative pathway at nearly every opportunity. The researchers then derived a preference score from these rejection patterns and found that it closely tracked the traditional choice-based preference score measured in standard simultaneous tests. The sequential measure was typically somewhat less extreme than the simultaneous one, but the overall pattern—the strong attraction to predictive information even at the cost of food—was unmistakably preserved.

But the story took a decisive turn when the researchers introduced a temporal cost to rejection. In this second phase, rejecting an option no longer came for free. Instead, each rejection incurred a fixed delay, and critically, the magnitude of that delay was adjusted across sessions based on each individual pigeon’s rejection rate. When the cost of rejection was zero or minimal, pigeons continued to reject the non-informative option and pursue the informative one. However, as the rejection delay grew longer, a clear shift emerged: pigeons increasingly accepted the non-informative option rather than paying the temporal price to search for the informative alternative.

The asymmetry in how the two options responded to cost is arguably the study’s most theoretically revealing finding. Rejection of the informative option remained rare throughout and was largely insensitive to the size of the delay. Pigeons accepted the informative option nearly unconditionally, whether rejection was free or expensive. Rejection of the non-informative option, by contrast, was highly sensitive to cost, declining steadily as the delay increased. This dissociation suggests that the two options were not being processed through a single, unitary valuation mechanism, but rather that their acceptance and rejection were governed by different underlying considerations.

The researchers interpret these patterns through the lens of rate-based foraging models, which hold that animals evaluate options not in isolation but in terms of their net contribution to overall feeding rate. Under this framework, the value of rejecting an option depends on the opportunity cost of doing so—the amount of food and time given up by postponing or forgoing the current offer. When rejection is free, the opportunity cost is low, and pigeons can indulge their strong preference for information without penalty. When rejection carries a delay, the opportunity cost rises, and the calculus shifts. At some point, the temporal price of searching for the informative option outweighs its subjective value, and the pigeon opts to simply accept the non-informative offer in front of it.

These findings carry significant implications for how scientists understand the suboptimal choice phenomenon. The persistence of information preference in a sequential search task confirms that the effect is not a quirk of simultaneous choice architecture but a deeper feature of how pigeons value predictive stimuli. At the same time, the sensitivity of rejection behavior to temporal cost demonstrates that the preference is not absolute: it flexes when the ecological price of information-seeking changes. This dual character—robust yet cost-sensitive—paints a more nuanced picture than either pure information-seeking or pure reward maximization alone would predict.

The work also connects to broader questions in behavioral ecology and foraging theory, where search-and-reject decisions are fundamental to how animals navigate patchy environments. Natural foragers rarely encounter all options simultaneously; they encounter opportunities sequentially and must decide at each step whether to exploit what is available or continue searching. By showing that the well-documented suboptimal choice effect extends into this ecologically realistic framework—and that it is modulated by the costs inherent in sequential search—the study bridges laboratory paradigms and natural foraging contexts. The research, supported by the Portuguese Foundation for Science and Technology and conducted with ethical approval from UCLA’s Institutional Animal Care and Use Committee, demonstrates that the tension between information and reward is not confined to the simultaneous choice apparatus but shapes decisions in the very sequential structures that evolution built.

The pigeons tested in the study were Columba livia, the same species that has served as the workhorse of comparative cognition for more than a century. Their long history in laboratory learning research makes them an ideal population for probing how valuation mechanisms respond to changes in task structure, since decades of prior work on timing, conditioning, and choice provide a rich theoretical backdrop against which new results can be interpreted.

The study moved through the peer-review process with notable speed. The manuscript was received in February 2026, revised in early June, accepted in late August, and published online on 10 September 2026 as an open access article, meaning that anyone can read, download, and share the full text under a Creative Commons Attribution 4.0 license. The authors note that the early-release version is citable and carries a permanent DOI, though it remains subject to editorial edits before the final Version of Record replaces it.

Financial and institutional support for the work came from several sources. The Portuguese Foundation for Science and Technology provided funding through multiple grants, and Marco Vasconcelos received a Fulbright Grant for Scholars and Researchers for the 2022/2023 academic year, supported by the Luso-American Development Foundation. The collaboration itself spans two countries, with authors affiliated with the William James Center for Research at the University of Aveiro in Portugal and the Department of Psychology at the University of California, Los Angeles.

All procedures involving the animals were reviewed and approved by UCLA’s Institutional Animal Care and Use Committee, and the authors state that the work was conducted in strict accordance with national and international guidelines for the care and use of laboratory animals. The authors also declare that they have no conflicts of interest.

The article is indexed under keywords that signal its interdisciplinary reach, including suboptimal choice, sequential choice, accept-reject decisions, foraging behavior, and Columba livia. Springer’s related-subject listings connect it to topics ranging from animal cognition and decision making to behavioral economics and operations research, reflecting the fact that questions about how organisms weigh information against reward resonate well beyond comparative psychology.

For readers who wish to examine the full experimental details, the article is available at its publisher’s site under DOI 10.1007/s10071-026-02101-0, with correspondence directed to Marco Vasconcelos. Because the paper is open access, the complete methods, results, and analyses can be consulted directly without subscription or payment.

Subject of Research: From choice to search: suboptimal preferences in sequential accept-reject decisions

Article Title: From choice to search: suboptimal preferences in sequential accept-reject decisions

Article References: Vasconcelos, M., Kirkman, C., Blaisdell, A. P., & Machado, A. (2026). From choice to search: suboptimal preferences in sequential accept-reject decisions. Animal Cognition. https://doi.org/10.1007/s10071-026-02101-0

Image Credits: AI Generated

DOI: 10.1007/s10071-026-02101-0

Keywords: choice, search, suboptimal, preferences, sequential, accept-reject, decisions, scientific research