The research, led by JiaMing Bai with colleagues Xiao Lv, ChenJiang Shi, Yu Zhang, and Qi Zhao, addresses a persistent weakness in metaheuristic optimization: premature convergence. Classical swarm algorithms, including the wolf pack algorithm that inspired the new work, tend to cluster their candidate solutions around an early promising region of the search space. Once that happens, the population loses diversity, the search stagnates, and the algorithm settles for a mediocre assignment schedule even when better ones exist. For time-critical military or disaster-response reconnaissance missions, that shortfall translates directly into missed targets or wasted flight hours.

The authors’ central design choice is to strip the wolf metaphor down to a structural scaffold while replacing its stochastic search behavior with more principled mechanisms. The wolf pack stages — scouting, calling, and besieging prey — remain as the framework that organizes how candidate solutions explore and exploit the landscape. But the random walks that drive movement between solutions are redrawn from a Lévy flight distribution, a heavy-tailed probability law in which occasional extremely long jumps punctuate many short, local steps. This pattern, observed in the foraging paths of real animals from albatrosses to deer, is mathematically attractive for optimization because it balances dense local sampling with rare long-range escapes that can carry the search out of a local optimum entirely.

Adaptivity is the second pillar of the method. Rather than fixing the step-size scale or the balance between exploration and exploitation for the entire run, the algorithm modulates these parameters in response to the evolving state of the population. Early in the search, when solutions are scattered, the algorithm favors wide-ranging Lévy jumps to map the terrain. As convergence progresses and the swarm tightens around promising regions, the step lengths shrink and the search shifts toward fine-grained refinement. This dynamic scheduling guards against the twin failure modes of metaheuristics: wandering aimlessly when exploration dominates, and locking in prematurely when exploitation takes over.

To evaluate the approach, the team benchmarked it against ten competing algorithms — a lineup that includes established metaheuristics such as particle swarm optimization, grey wolf optimizer, and whale optimization algorithm, alongside the original wolf pack algorithm and a probability-threshold variant labeled EWPA-PT. The testbed consisted of 300 randomly generated instances of the multi-UAV task allocation problem, organized into three scales of 100 instances each. The largest problems involved 50 UAVs and 100 tasks, a scale at which exact combinatorial methods become computationally intractable and heuristics are the only practical option.

The results were unambiguous. Across all 300 instances, the adaptive Lévy wolf pack algorithm achieved the best mean objective values, and nonparametric statistical tests confirmed that the improvements were significant rather than artifacts of random variation. On the largest benchmark class, the method reduced the mean objective by roughly 8.3 percent compared with the strongest competing algorithm, a margin the authors describe as practically meaningful for mission planning. Perhaps more telling was the consistency: the new algorithm showed the lowest variance in solution quality across independent runs, indicating that its performance does not hinge on a lucky random seed.

Runtime told a similar story. The adaptive variant completed the large-scale instances in about 4.6 seconds on average, roughly 20 percent faster than the original wolf pack algorithm and an order of magnitude quicker than several competitors that exceeded 40 seconds per instance. The authors attribute this efficiency to the streamlined search dynamics — fewer wasted evaluations spent revisiting already-explored regions — and to the adaptive mechanism’s ability to terminate refinement once improvements plateau.

The study also probes the algorithm’s internal behavior through ablation-style analysis. Removing the adaptive component and fixing parameters at static values degraded both solution quality and convergence speed, confirming that the dynamic tuning is not decorative but load-bearing. Similarly, replacing the Lévy flight with Gaussian random walks produced noticeably worse results on the hardest instances, supporting the hypothesis that heavy-tailed jumps are what allow the search to escape deep local optima in the rugged assignment landscape.

Beyond the benchmarks, the authors demonstrate the method on a realistic reconnaissance scenario with geographically distributed targets and UAVs constrained by range and endurance. The resulting schedules respect all operational constraints while achieving near-optimal coverage scores, suggesting the approach could be integrated into real mission-planning software. The framework is also generic: because the wolf pack structure serves purely as an organizational scaffold, the same Lévy-driven, adaptively tuned search could be applied to other combinatorial problems such as vehicle routing, satellite scheduling, or sensor placement.

The work arrives amid a broader debate about the proliferation of nature-inspired metaheuristics, many of which critics argue differ from one another mainly in metaphor rather than mechanism. By explicitly identifying which components — the Lévy jumps, the adaptive parameter control, the structural stages — contribute measurable performance, this study offers a template for more rigorous algorithm design. For the growing community deploying drone swarms in agriculture, surveillance, and emergency response, the message is practical: how you randomize your search matters, and tuning that randomness to the problem’s evolving structure can turn a sluggish optimizer into a mission-ready planner.

Subject of Research: Swarm-intelligence optimization for multi-UAV reconnaissance task allocation

Article Title: Adaptive lévy wolf pack algorithm for multi-UAV cooperative reconnaissance task allocation

Article References: Bai, J., Lv, X., Shi, C., Zhang, Y., & Zhao, Q. (2026). Adaptive lévy wolf pack algorithm for multi-UAV cooperative reconnaissance task allocation. Cluster Computing, 29(15), 864. https://doi.org/10.1007/s10586-026-06629-1

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06629-1

Keywords: UAV swarms, task allocation, swarm intelligence, Lévy flight, wolf pack algorithm, metaheuristics, combinatorial optimization, reconnaissance, adaptive algorithms, drone missions, Adaptive, wolf

Muhammad Zubair and Syed Khaliq Shah of Southeast University in Nanjing tackled what they call the integrated rebalancing and faulty bike recycling problem. Instead of sending trucks out on separate missions—one to redistribute bikes, another to collect broken ones—their framework folds both tasks into a single vehicle routing model. Each service truck carries a partitioned cargo hold: a main compartment for rebalancing bikes and a dedicated section for faulty ones. This partitioned capacity design means a vehicle can deliver bikes to deficit stations, collect surplus bikes from oversupplied ones, and scoop up damaged bicycles along the way, all in one tour.

The mathematics behind the model is demanding. The researchers formulated it as a variant of the pickup and delivery problem, with an objective function that balances three competing costs: user dissatisfaction when riders cannot find a bike, operational expenses including travel and loading time, and the handling cost of faulty bicycles. Loading or unloading an operational bike takes about one minute, while a damaged bike requires roughly two minutes of handling time—values drawn from real observations of Nanjing operators. Vehicles must complete their tours within a six-hour shift, and travel times are adjusted by traffic congestion factors derived from Baidu Maps data, ranging from free-flow conditions to heavy congestion.

Exact optimization tools like IBM’s CPLEX solver can crack such models, but only at small scale. The team first validated their formulation on 38 real stations near Southeast University, solving it to near-perfect optimality with percentage gaps as low as 0.02 percent. In 35 of the 38 stations, rebalanced inventory aligned almost exactly with forecast demand, and all faulty bikes were successfully routed back to the depot for repair or recycling. But real-world networks dwarf this testbed: Shanghai operators manage more than 50,000 dockless bikes, and New York’s CitiBike runs over 1,000 stations. Exact mixed-integer programming becomes computationally intractable beyond roughly 50 stations, which is where the researchers’ metaheuristic enters.

Their solution combines a genetic algorithm with adaptive large neighborhood search, wrapped in a cluster-first, route-second decomposition. The genetic algorithm evolves populations of candidate routes using tournament selection, order crossover, and mutation operators, while preserving the best solutions through elitism. The ALNS component then intensifies the search by destroying parts of a solution—removing stations at random, targeting the worst-performing segments, or dismantling entire clusters—and repairing them with greedy, regret, or cheapest insertion strategies. An adaptive weighting mechanism learns which destroy-and-repair combinations perform best, and a simulated annealing acceptance criterion decides whether to embrace worse solutions temporarily to escape local optima.

The clustering stage is where the approach departs most sharply from prior work. Rather than dividing a city into fixed geographic zones, the dynamic clustering procedure jointly optimizes the number of clusters, their centroids, and station assignments based on expected inter-cluster traversal costs. This captures a fundamental trade-off: more clusters shrink the distance vehicles travel within each zone but inflate the cost of moving between zones. In experiments on a 200-station instance, increasing clusters from 3 to 20 cut intra-cluster travel by 65 percent while inflating inter-cluster travel by more than 700 percent. The sweet spot landed at 11 clusters, where total routing distance fell to 362 kilometers—a 26.1 percent improvement over three clusters and a 21.8 percent improvement over twenty.

The performance numbers are striking. On 200-station synthetic instances, the hybrid algorithm produced solutions within 3.2 percent of best-known values in about 15 minutes, a fourfold scale improvement over the most comparable integrated studies, which stall at 50 to 100 stations. Against 50-station benchmarks, the method matched exact solvers with gaps below 0.5 percent while running 3.75 times faster. Dynamic clustering itself shaved 6.5 percent off routing distance compared with static, geography-only approaches. Meanwhile, folding faulty bike collection into rebalancing routes added just 8 to 12 percent computational overhead—yet eliminated the separate collection trips entirely, cutting total vehicle distance by 15 to 25 percent compared with running the two operations sequentially.

The practical implications are substantial. For a 200-station city, the researchers estimate daily savings of 50 to 75 kilometers of vehicle travel, translating to roughly 10 to 15 liters of fuel, two to three hours of driver time, and annual cost reductions of 15,000 to 25,000 US dollars per vehicle. The recycling loop itself follows a four-stage pipeline: IoT sensors and GPS monitoring flag faulty bikes based on battery status, signal strength, failed QR scans, and idle periods exceeding seven days; trucks with partitioned holds collect them during routine rebalancing; capacity constraints keep the two cargo types physically separated throughout the route; and the depot sorts arrivals into immediate repair, scheduled maintenance, or parts recovery.

Empirical research from major Chinese cities underscores why this integration matters. Faulty bike rates can exceed 2 percent during peak usage, and delayed collection measurably damages user satisfaction and safety. Earlier integrated models relied on exact optimization that buckled beyond 50 stations, and most treated vehicle capacity as a single undifferentiated constraint, implicitly assuming sequential loading and unloading. The new framework’s partitioned capacity model and jointly optimized clustering address both shortcomings, and the authors argue the approach could extend beyond bike sharing to fleet management, ride-sharing, and last-mile delivery, where decomposition is often treated as a necessary evil rather than an optimization opportunity.

Limitations remain. The model assumes deterministic demand within a single operational period and a homogeneous fleet, simplifications that real cities may strain against. The researchers point to several frontiers: incorporating stochastic demand through robust optimization, extending to multi-depot networks, adapting the algorithm for real-time rebalancing as conditions change mid-shift, integrating dynamic pricing to encourage users to self-balance the system, and pairing the optimizer with machine learning demand forecasts. For now, the study offers something operators have lacked: a computationally feasible way to keep dockless fleets balanced, functional, and profitable at the scale modern cities actually demand.

Subject of Research: Integrated rebalancing and faulty bike recycling optimization in dockless bike-sharing systems using a hybrid metaheuristic with dynamic clustering

Article Title: Integrated rebalancing and faulty bike recycling in dockless bike-sharing systems: a hybrid metaheuristic approach with dynamic clustering

Article References: Zubair, M., & Shah, S. K. (2026). Integrated rebalancing and faulty bike recycling in dockless bike-sharing systems: a hybrid metaheuristic approach with dynamic clustering. Frontiers in Future Transportation, 7, Article 1919819. https://doi.org/10.3389/ffutr.2026.1919819

Image Credits: AI Generated

DOI: 10.3389/ffutr.2026.1919819

Keywords: bike-sharing, dockless bikes, vehicle routing, metaheuristics, genetic algorithm, adaptive large neighborhood search, dynamic clustering, faulty bike recycling, urban mobility, optimization, Nanjing, transportation systems

The researchers propose a hybrid framework they call WOA-EEMD-CNN-LSTM, a pipeline in which three distinct computational ideas are chained together, each one addressing a specific weakness of the previous stage. The first stage uses ensemble empirical mode decomposition, or EEMD, a noise-assisted signal-processing technique first introduced by Wu and Huang in 2009 as an extension of Huang’s original empirical mode decomposition from 1998. EEMD takes the messy, jagged oxygen time series and breaks it down into a set of intrinsic mode functions, each capturing oscillations at a particular scale, plus a residual trend. Instead of asking one model to learn everything at once, the decomposition hands the downstream learner a family of simpler, better-behaved sub-signals.

But EEMD has a well-known catch: its quality depends critically on parameters, notably the amplitude of the white noise injected into the ensemble and the number of ensemble trials. Set them poorly and the decomposition either fails to cancel out the noise or becomes computationally prohibitive. This is where the whale enters the story. The Whale Optimization Algorithm, introduced by Mirjalili and Lewis in 2016, is a metaheuristic inspired by the bubble-net hunting strategy of humpback whales, in which the animals swim in shrinking spiral patterns to corral prey. In the new framework, WOA is used to search the parameter space and tune the key EEMD settings automatically. The authors describe this as entropy-guided parameter adaptation, meaning the optimization is steered by information-theoretic measures of how well the decomposition separates the signal into meaningful components.

Once the oxygen series has been decomposed into its multi-scale components, the framework deploys a combined convolutional and recurrent neural network to forecast each component individually. The convolutional layer, following the principles established by LeCun and colleagues in their landmark 1998 work on gradient-based learning, acts as a local feature detector, scanning short windows of each sub-series for characteristic patterns. The long short-term memory network, the celebrated 1997 architecture of Hochreiter and Schmidhuber, then models the long-term temporal dependencies that determine how each component evolves. Because the components are largely decoupled from one another after decomposition, each CNN-LSTM pair can specialize, learning the dynamics of a slow trend without being confused by high-frequency jitter, and vice versa. Finally, the component predictions are aggregated back together using timestamp alignment, reconstructing a single coherent forecast of dissolved oxygen.

The evaluation is arguably the most rigorous part of the paper. The authors tested the framework on two very different datasets: a four-hour-interval monitoring series from the Xiaojiang estuary in China and a publicly available dataset known as the Refuge dataset, hosted on Kaggle. To keep the comparison honest, they fixed the random seed at 123 in what they call a locked seed evaluation, ensuring that reported numbers reflect the method itself rather than a lucky draw of random initializations. On the Xiaojiang series, the WOA-EEMD-CNN-LSTM model achieved a root mean square error of 0.32867 and a coefficient of determination, or R-squared, of 0.95070, meaning the model explains roughly ninety-five percent of the variance in the oxygen signal. On the Refuge dataset, it recorded an RMSE of 0.28874 and an R-squared of 0.93429.

Those figures place the framework in respectable territory for dissolved-oxygen forecasting, a field where earlier machine learning efforts have produced a wide spread of results. The literature the authors survey reads like a catalogue of the field’s evolution: artificial neural networks and grey systems models for river classification, k-nearest-neighbor approaches paired with particle swarm-optimized LSTMs, variational mode decomposition combined with sparrow search algorithms and support vector machines on the Yangtze, and EEMD paired with LSTM networks for the Xiaofu River. Hybrid decomposition-plus-deep-learning pipelines have also proven themselves far beyond water quality, including hybrid EMD-LSTM models for non-stationary wave prediction in offshore China and EEMD-CNN-LSTM hybrids for crude oil futures. What distinguishes the new work is the explicit optimization of the decomposition stage itself, rather than leaving those parameters to trial and error.

Beyond raw accuracy, the authors emphasize a set of design principles that make the workflow unusually transparent for a deep learning study. The framework provides entropy-guided parameter adaptation so that the decomposition is not a black box bolted onto the front of the network; a causal multi-scale representation so that no information from the future leaks into the past; component-wise learning so that the network’s burden is divided into tractable sub-problems; and timestamp-aligned aggregation so that the reconstructed forecast lines up exactly with the physical timeline of the measurements. Perhaps most notably, they highlight an auditable, leakage-controlled evaluation process, an issue that has dogged time-series machine learning generally, where subtle data leakage between training and test sets can inflate reported accuracy dramatically.

That emphasis on reproducibility matters because dissolved-oxygen forecasting is not an academic exercise. Oxygen levels function as a primary indicator for classifying river health and assessing urban impacts, as established in classic environmental monitoring work dating back to 2007. Machine learning models of oxygen concentration have been applied with dimensionality reduction techniques and with increasingly sophisticated architectures, but deployment decisions, such as issuing an aeration alert at an aquaculture facility or triggering an investigation into an illegal discharge, require forecasts that regulators can trust and verify. A framework that documents its random seeds, controls for data leakage, and decomposes its predictions into interpretable components offers an audit trail that a single opaque network trained on raw sensor data simply cannot provide.

The authors are careful about the limits of what they have demonstrated. Their evaluation rests on two sites, one estuarine and one from a public repository, and they explicitly position the current locked-seed results as a basis for later multi-seed and multi-site validation rather than a final verdict. This matters because time-series models can be sensitive to the specific regime they are trained on: an estuary dominated by tidal mixing behaves differently from a reservoir stratified by season. The Xiaojiang dataset is available from the corresponding author on reasonable request, while the Refuge dataset is fully public, which at least gives independent researchers a path to test the pipeline for themselves. The authors also report that no specific financial support was received for the study and declare no competing interests.

Still, the direction of travel is clear. As climate change intensifies the swings between flood-driven nutrient pulses and drought-driven stagnation, the oxygen dynamics of rivers and estuaries are becoming harder to anticipate with simple statistical models, and the cost of a missed hypoxic event keeps rising. Hybrid pipelines that combine physics-aware signal decomposition, metaheuristic optimization, and deep temporal learning represent a pragmatic middle path: they exploit the strengths of modern artificial intelligence while keeping the signal-processing foundations visible and tunable. If follow-up studies confirm that WOA-optimized decomposition holds up across many sites and seasons, the framework described by Xie and Liang could become a template for the next generation of early-warning systems, giving water managers a few precious hours of foresight before the oxygen disappears and the fish follow.

Subject of Research: Machine learning prediction of dissolved oxygen in water-quality monitoring time series

Article Title: A hybrid framework for water-quality prediction using WOA-optimized EEMD and CNN-LSTM

Article References: Xie, X., & Liang, J. (2026). A hybrid framework for water-quality prediction using WOA-optimized EEMD and CNN-LSTM. Environmental Monitoring and Assessment, 198(11), Article 1189. https://doi.org/10.1007/s10661-026-16043-0

Image Credits: AI Generated

DOI: 10.1007/s10661-026-16043-0

Keywords: dissolved oxygen, water quality, deep learning, CNN-LSTM, ensemble empirical mode decomposition, whale optimization algorithm, time-series forecasting, signal decomposition, environmental monitoring, metaheuristics, estuary, machine learning

The motivation behind CSOMA stems from a recurring bottleneck in statistical modeling, data science, and experimental design. In many realistic applications, the objective function a researcher needs to minimize is nonlinear, nonconvex, nonsmooth, simulation-based, or available only through a numerical approximation. These awkward features appear constantly in model-based optimal experimental design, in likelihood-based estimation with complex dependence structures, in regression with distribution-valued responses, in design problems posed on curved manifolds, and in continuous reformulations of combinatorial tasks. When derivatives are unreliable or algebraically intractable, gradient-based solvers stumble, and researchers are often forced to write one-off scripts tied to a single example. CSOMA addresses this gap by packaging a derivative-free optimizer behind a deliberately narrow interface: users write a scalar objective function, specify lower and upper bounds, choose a swarm size and iteration budget, and call a single function named csoma().

The algorithmic heart of the toolkit is an evolutionary twist on particle-swarm thinking. Conventional particle swarm optimizers track personal best positions and a global best, updating every particle toward these remembered landmarks. CSO-MA takes a leaner approach: at each iteration, the swarm is randomly permuted and paired off into head-to-head competitions. In each pair, the particle with the lower objective value is declared the winner and stays put, while the loser is updated using a velocity rule that blends three influences. The first term carries over the loser’s previous velocity, preserving momentum. The second term moves the loser toward the winner of its own pairing, extracting local order information from objective evaluations without any gradient calculation. The third term pulls the loser toward the empirical center of the entire swarm, with a tunable parameter phi controlling the strength of this population-level attraction. If the swarm contains an odd number of particles, one simply sits out the round.

After the loser updates, CSOMA adds a distinctive step that gives the method its name: the mutated-agent mechanism. One updated loser is selected at random, one of its coordinates is chosen, and that coordinate is reset to the corresponding lower bound, upper bound, or interval midpoint. This deliberately simple mutation probes boundary regions that are common in bounded design parameterizations and helps the stochastic search recover from premature concentration, where all particles have clustered in one region of the search space. A clipping step then ensures every position remains inside the user-specified box, so feasibility is enforced by construction. The developers note that each pairwise competition extracts useful ordering information from objective evaluations alone, making the method a natural match for design criteria whose values can be computed through information matrices, determinants, likelihood approximations, or simulation summaries, but whose derivatives may be numerically unstable or impossible to write down cleanly.

The software architecture keeps model-specific code strictly outside the optimizer. The core function, src/csoma.m, uses only base MATLAB operations, requires no compilation, and depends on no external data. A setup helper exposes the package folders, while bundled examples demonstrate how statistical criteria can be encoded as function handles and optimized. The examples span an impressive range: finding locally D-optimal designs for logistic regression, constructing Bayesian designs for nonlinear mixed-effects models of HIV dynamics, optimizing Gaussian copula designs and maximum pseudo-likelihood estimates, fitting Wasserstein regressions for distribution-valued data, designing experiments on the surface of a sphere using Riemannian geometry, and even solving a ten-city traveling salesman problem by sorting continuous random keys. Heavier optional examples, such as the Bayesian HIV workflow, require additional MATLAB toolboxes, but the core optimizer and fast replication checks run on a standard laptop in seconds.

Reproducibility is where the package makes its most forceful statement. The repository includes a reviewer-facing script, replication/run_all.m, that executes the fast workflows with fixed random seeds, and a release-verification wrapper, replication/verify_round2.m, that records the full MATLAB version, operating-system details, installed products, and exact Git commit before running the automated suites. Numerical success is defined by explicit assertions rather than visual inspection of console output: a run passes only when every workflow satisfies finite-value, feasibility, normalization, and problem-specific criteria. The final retained verification was executed on September 18, 2026, using MATLAB R2025b Update 7 on GitHub-hosted Ubuntu 24.04, with the public v0.1.5 tag resolving to a specific forty-character commit and a clean worktree. Deterministic seeds make the stochastic workflow auditable, and the core smoke test requires two seeded optimizer calls to produce exactly equal results.

The flagship statistical application is illustrated by a two-factor logistic regression design problem on the square from minus one to one in each coordinate. Assuming a nominal parameter vector of (2, 3, minus 1), CSOMA searches continuously for optimal support points and their weights under the D-optimality criterion, which seeks designs that maximize the determinant of the Fisher information matrix. Crucially, the computed design can be checked against the classical equivalence theorem of Kiefer and Wolfowitz, which characterizes global optimality through a directional-derivative inequality. Under the centered convention used in the paper, the sensitivity function must be no greater than zero throughout the design space and must equal zero at the support points of an optimal design. Evaluating the sensitivity on a fine 201-by-201 grid, the authors found a maximum centered sensitivity of just 1.62 times ten to the minus six, essentially zero, confirming that the swarm-found design is D-optimal for practical purposes.

The Wasserstein regression example reveals the geometric depth the toolkit can reach. Distribution-valued responses, where each observation is an entire probability distribution rather than a single number, can be analyzed through the geometry of optimal transport. In one dimension, the squared Wasserstein distance between two distributions equals the integral of the squared difference of their quantile functions, and the bundled example exploits this by fitting a normal-quantile model and minimizing mean squared quantile error over an eighty-point probability grid. The paper visualizes this as linear motion in quantile coordinates, with the monotone transport map and its associated displacement geodesics providing an elegant interpretation of the loss being minimized. Meanwhile, the Gaussian copula examples show that the same optimizer layer works both upstream, choosing informative experimental settings by maximizing information, and downstream, estimating dependence parameters when a closed-form optimizer is inconvenient.

The developers are candid about limitations. As a stochastic metaheuristic, CSOMA guarantees no recovery of a global optimum in a finite number of iterations, and performance depends on swarm size, iteration budget, random seed, variable scaling, and the numerical stability of the user-supplied objective. The interface directly supports only box constraints; other constraints must be handled through transformations or penalties, and objective functions must return finite scalar values, with infeasible settings managed through penalties or large finite values. The team recommends that users examine stability across multiple seeds and increase the swarm size or iteration budget when results remain unstable. When reliable derivatives are available, mature gradient-based solvers may still be preferable, and the authors explicitly frame CSOMA as a narrow, inspectable tool rather than a replacement for comprehensive optimization suites.

Nevertheless, the potential impact on day-to-day research practice is considerable. By turning a method that previously existed mainly in methodological papers into an accessible, MIT-licensed tool distributed on GitHub and the MATLAB File Exchange, the team hopes to lower the cost of reproducible derivative-free optimization across statistics and data science. The immediate beneficiaries are experimental designers working with nonlinear models, integer run sizes, and simulation-based utilities, where CSO-MA has already demonstrated practical efficiency gains, including applications in developmental toxicology using sea urchin embryos. But the broader pattern matters more: any researcher who can formulate a scalar loss can now attach it to a shared, versioned optimizer, rerun published examples, compare alternative criteria, and hand colleagues a workflow they can inspect and execute. In a field where irreproducible optimization scripts have long been the quiet weak point of published analyses, that kind of transparency may prove to be the most valuable result of all.

Subject of Research: A reusable MATLAB toolkit for derivative-free optimization in optimal experimental design

Article Title: CSOMA : A reusable and reproducible MATLAB toolkit for derivative-free optimization in optimal experimental design

Article References: Cui, E. H., Zhang, Z., Pan, Y. E., & Wong, W. K. (2026). CSOMA: A reusable and reproducible MATLAB toolkit for derivative-free optimization in optimal experimental design. SoftwareX, 36, Article 103121. https://doi.org/10.1016/j.softx.2026.103121

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103121

Keywords: CSOMA, derivative-free optimization, optimal experimental design, competitive swarm optimizer, MATLAB, reproducibility, D-optimal designs, metaheuristics, Wasserstein regression, copula models, statistical computing, open-source software

The work, authored by Weiwei Zhao of Anhui Sanlian University, addresses a stubborn gap in document-level neural machine translation, or DNMT. Conventional Transformers excel at modeling sequences within a sentence, yet their standard formulation does not explicitly carry evidence across sentence boundaries. That limitation is especially consequential for the English-Chinese language pair. English tends to mark person, number, tense, and logical relations overtly, whereas Chinese relies heavily on topic continuity, zero pronouns, ellipsis, lexical repetition, and context-sensitive aspect. A system blind to document context may pick the wrong antecedent for a pronoun, translate the same technical term two different ways, or drop an implicit participant that Chinese grammar leaves unstated.

Previous context-aware models have chipped away at the problem, but the study argues that most share structural weaknesses. Many rely on fixed context windows: too short, and they miss distant antecedents; too long, and they flood the model with noise, omissions, and attention dilution. Others compress local syntax, entity evidence, and rhetorical relations into a single undifferentiated representation, even though these phenomena demand different kinds of attention. Recent evaluations of large language models add a cautionary note: high aggregate scores do not guarantee that a model actually uses the decisive contextual cue, particularly for pronoun translation. The new framework, called DPI-RFO with a Context-Aware Transformer (CAT), is designed around the idea that context representation and model configuration should be solved together rather than as two independent stages.

Architecturally, CAT keeps three kinds of information in separate attention pathways. Sentence self-attention handles local syntax and meaning, exactly as a standard Transformer would. A cross-sentence context pathway lets the current sentence query representations of surrounding document text, retrieving lexical and entity evidence from beyond its own boundaries. A third, discourse-aware pathway incorporates an additive bias derived from an English discourse parser, which scores temporal, causal, contrastive, and elaborative relations between sentence pairs. The parser’s output is treated as a soft prior rather than a hard constraint: relation scores are mapped to learned embeddings, scaled by parser confidence, and added to attention logits before the softmax. If the parser errs, the mistake only shifts attention slightly and can be suppressed downstream.

The crucial glue is token-wise gated fusion. The three pathway outputs are concatenated, projected, and passed through a sigmoid gate that decides, for every individual token, how much fused contextual information versus the residual sentence-level representation should flow forward. This mechanism lets the model lean heavily on context when translating an ambiguous pronoun while ignoring it when local syntax alone suffices. The study’s attention heatmaps illustrate the effect qualitatively: standard self-attention concentrates on diagonal, within-sentence patterns, while CAT attention activates strongly across sentence boundaries and the discourse-enhanced variant highlights regions tied to coreference and temporal relations.

The second half of the framework is where the paper gets genuinely unusual. Choosing a context-aware architecture involves a tangled web of coupled decisions: context-window length, number of attention heads, encoder and decoder depth, hidden dimensions, dropout, learning rate, batch size, and more. Manual tuning and small grid searches cannot adequately explore this mixed discrete-continuous space. The author turns to a swarm-based metaheuristic called Rüppell’s Fox Optimization, inspired by the encirclement and pouncing behavior of the Saharan fox, and then modifies it. The enhanced version, DPI-RFO, adds a cosine-decaying perturbation schedule that keeps exploration broad early and tightens it later, plus a stagnation-triggered Lévy-flight mechanism that fires occasional long-distance jumps when the population of candidate configurations stops improving.

The rationale is empirical as much as theoretical. A monotonically shrinking search radius can collapse a population around a configuration that looks promising during partial training but underperforms after convergence, a known trap in neural hyperparameter optimization. Under a matched budget of 5,600 candidate evaluations, with every candidate receiving identical 20,000-step training fidelity, DPI-RFO reached the target fitness with 2,780 evaluations on average, roughly 30.5 percent fewer than baseline RFO and fewer than random search, Bayesian optimization, Hyperband, BOHB, population-based training, and other swarm methods including particle swarm, grey wolf, and whale optimization. The search also jointly minimized a composite objective balancing validation BLEU against normalized computational cost, discouraging needlessly large architectures.

The evaluation was unusually thorough. Six English-Chinese domains were tested: news, spoken language, subtitles, legal, medical, and literary texts, drawn from corpora including News Commentary, TED Talks, OpenSubtitles, Patent-ZH, a biomedical parallel corpus, and webnovel data, totaling roughly 13.77 million aligned sentence pairs across about 457,300 document units. All results were averaged over five independent seeds, and all baselines were retrained under identical preprocessing, tokenization, decoding, and evaluation protocols. On the news test set, the full system achieved 50.8 BLEU, 49.5 document-level BLEU, and 67.8 Coherence-BLEU, a composite metric blending translation accuracy with discourse-consistency scoring. Gains over the strongest baseline, a selective-attention context model, ranged from 1.8 to 2.5 BLEU points across all six domains, with the largest improvement in the medical domain.

Ablation analysis clarified where the credit lies. Removing the entire CAT block dropped BLEU from 50.8 to 44.3, confirming that the discourse-aware representation is the primary source of quality gains, while removing only the dynamic perturbation reduced BLEU to 47.9, showing the optimizer’s complementary contribution to search efficiency and stability. Fine-grained metrics told a consistent story: the system improved lexical cohesion, anaphora resolution, discourse-connective accuracy, and entity consistency, with the biggest jumps in connective F1 and entity consistency. Paired bootstrap testing over 12,485 documents confirmed the improvements were statistically significant, and a blinded human evaluation by three professional translators, covering 200 documents and 12,000 dimension-level ratings, found the highest scores on every dimension, with the largest advantages in pronoun resolution and terminology consistency.

The study is candid about costs and limits. The selected model carries 495 million parameters and processes 155 sentences per second on an A100 GPU in FP16, a modest overhead relative to its quality gains but better suited to batch document translation than latency-critical serving. The one-time search consumed roughly 2,920 GPU-hours, with final training adding 288 more. The framework also assumes reasonably preserved document boundaries and usable discourse cues, and all experiments were confined to English-Chinese, leaving generalization to other language pairs untested. Still, the paper’s central demonstration stands: when a translation model is built to reason about discourse explicitly, and when its architecture is tuned by a search procedure that refuses to stagnate, documents come out not just more accurate sentence by sentence, but more coherent as wholes.

Subject of Research: Document-level English-to-Chinese neural machine translation using a context-aware transformer and metaheuristic architecture search

Article Title: Dynamically perturbed Ruppell Fox optimization with a context aware transformer for coherent document level English to Chinese translation

Article References: Zhao, W. (2026). Dynamically perturbed Ruppell Fox optimization with a context aware transformer for coherent document level English to Chinese translation. Discover Artificial Intelligence, 6(1), Article 1469. https://doi.org/10.1007/s44163-026-02407-5

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02407-5

Keywords: machine translation, document-level translation, English-Chinese translation, context-aware transformer, neural architecture search, metaheuristics, Rüppell's Fox optimization, discourse coherence, natural language processing, hyperparameter optimization, Lévy flight, anaphora resolution

The algorithm, named TSDE-SSA, stands for Two-Stage Differential Evolution with Squirrel Exploration. Differential evolution, first proposed by Rainer Storn and Kenneth Price in 1997, is a population-based method that evolves a swarm of candidate solutions by repeatedly mixing them together. It creates new candidates by taking the difference between two existing solutions, scaling that difference, and adding it to a third, a mutation scheme that lets the algorithm sense the contours of the search landscape. The method is prized for its simplicity and effectiveness, but like all metaheuristics it suffers from a well-known weakness: as problems grow more complex and higher-dimensional, classical differential evolution tends to converge prematurely, losing the population diversity it needs to keep exploring, and its local search ability degrades in high-dimensional spaces.

The central insight behind TSDE-SSA is that exploration and exploitation, the two competing forces of any search algorithm, are best served by different machinery at different times. Exploration means casting a wide net across the search space to locate promising regions; exploitation means drilling down into those regions with fine-grained steps to pin down the exact optimum. Most algorithms try to balance these forces continuously throughout a run, which can leave both half-done. The Chinese team instead divides the optimization process cleanly into two phases: an early exploration phase dedicated to global search, and a later exploitation phase devoted to local refinement.

In the exploration phase, TSDE-SSA combines the strengths of two classic differential evolution mutation strategies. The first, known as DE/rand/1, builds new candidates from randomly chosen population members, which preserves diversity and encourages broad sweeping of the search space. The second, DE/best/2, leans on the current best solution, which accelerates progress toward promising areas but risks herding the whole population into a single basin. By blending both strategies, the algorithm can sweep widely while still being pulled toward regions that look genuinely promising, swiftly identifying potential optimal solution regions before committing resources to them.

A further refinement in this early phase is a self-adaptive mechanism for historical parameters. Differential evolution’s behavior is governed by control parameters, most famously the scaling factor that weights the difference vectors. Fixed settings that work well for one problem can cripple performance on another, and decades of research have produced increasingly sophisticated adaptation schemes, from self-adapting parameters embedded in individuals to success-history adaptation as seen in the celebrated SHADE family of algorithms. TSDE-SSA adjusts its parameters based on their historical performance, allowing the algorithm to learn which settings serve it well as the run unfolds rather than forcing a human user to guess in advance.

Then comes the squirrel. The second phase of TSDE-SSA mimics the foraging behavior of the Squirrel Search Algorithm, a nature-inspired metaheuristic introduced by Jain and colleagues in 2019 that models how squirrels glide between trees in search of acorns, balancing the pull of nearby food sources against the risk of predators. In TSDE-SSA this gliding dynamics is integrated with differential evolution’s crossover operations, which recombine candidate solutions in a way that preserves useful building blocks. By dynamically adjusting parameters such as the scaling factor and the gliding constant during this later phase, the algorithm sharpens its local search accuracy while guarding against the premature convergence that plagues conventional differential evolution when the population collapses too early.

The authors subjected their algorithm to a demanding battery of tests. On the CEC2017 and CEC2020 benchmark suites, standardized collections of test functions used by the evolutionary computation community to compare optimization algorithms under controlled conditions, TSDE-SSA surpassed three fundamental baseline algorithms and five outstanding differential evolution variants in convergence speed, precision, and stability. Convergence speed measures how quickly the algorithm approaches a good solution, precision measures how close it gets to the true optimum, and stability measures how consistently it performs across repeated independent runs, a crucial property since these methods are stochastic and can produce different results each time they are invoked.

The comparison did not stop there. TSDE-SSA was also pitted against eight newly proposed metaheuristic algorithms, the kind of recent arrivals that populate the fast-moving swarm intelligence literature, including algorithms inspired by grey wolves, whales, Harris hawks, dung beetles, and hippopotamuses. Once again the two-stage hybrid came out ahead, suggesting that the structured division between exploration and exploitation, rather than any single exotic metaphor, is what drives the performance gain. Statistical rigor matters in this field, and the study follows established practice by using nonparametric statistical tests to compare algorithms, the methodology recommended for fairly assessing evolutionary and swarm-based methods.

Benchmarks alone do not convince engineers, so the team also deployed TSDE-SSA on five typical engineering design problems, the classic constrained optimization tasks that serve as proving grounds for new algorithms. The algorithm showed strong applicability and reliability across these real-world problems, demonstrating that its benchmark performance translates into practical utility. This matters because engineering optimization problems differ from clean benchmark functions in important ways: they carry constraints, mixed variable types, and objective landscapes shaped by physical laws rather than mathematical convenience. An algorithm that can navigate both worlds is far more valuable than one that excels only in the laboratory.

The work arrives amid a flourishing of hybrid differential evolution research, with recent years producing variants that incorporate opposition-based learning, Gaussian tail mutation, reinforcement learning-based strategy selection, competitive frameworks, and population reduction strategies, applied to everything from brain tumor detection and space mission trajectory design to photovoltaic parameter estimation and portfolio optimization. What distinguishes TSDE-SSA in this crowded field is its explicit two-stage architecture paired with the squirrel-inspired gliding mechanism, an approach the authors say effectively balances global and local search and offers an efficient and stable solution for complex optimization problems with broad application potential. As machine learning systems grow larger and engineering designs grow more intricate, the demand for optimizers that neither give up too early nor wander forever will only intensify, and this team’s two-stage squirrel suggests that sometimes the best way forward is to first look around, and only then pounce.

Subject of Research: A two-stage differential evolution algorithm combining squirrel search exploration with adaptive parameter control for complex optimization problems

Article Title: A two-stage differential evolution algorithm for collaborative squirrel exploration and its engineering application

Article References: Chen, H., Gao, Y., & Zhang, Y. (2026). A two-stage differential evolution algorithm for collaborative squirrel exploration and its engineering application. Cluster Computing, 29(12), Article 706. https://doi.org/10.1007/s10586-026-06516-9

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06516-9

Keywords: differential evolution, squirrel search algorithm, metaheuristics, global optimization, exploration and exploitation, adaptive parameters, CEC2017 benchmarks, CEC2020 benchmarks, engineering design, swarm intelligence, premature convergence, Cluster Computing

The original RIME algorithm belongs to the family of physics-based metaheuristics, computational techniques that borrow metaphors from natural phenomena to guide a population of candidate solutions through a search space. In RIME, each candidate solution is treated conceptually as a droplet that can crystallize into either soft rime or hard rime, two forms of ice that deposit differently on surfaces depending on environmental conditions. Soft rime corresponds to broad, exploratory movement across the search space, while hard rime corresponds to fine-grained refinement around promising regions. By balancing these two crystallization behaviors, RIME aims to avoid the twin failure modes of optimization: wandering forever without converging, and converging too quickly on a mediocre solution that merely looks good locally.

That balance, however, is exactly where the original algorithm struggled. According to the authors, RIME suffers from limited exploratory capability, slow convergence, and an imbalance between exploration and exploitation, the technical terms for searching widely versus digging deeply. In high-dimensional big data problems, these weaknesses compound: a search that explores too timidly may never reach the region containing the true optimum, while one that exploits too eagerly becomes trapped in local optima that are far from the global best. For feature selection tasks, where an algorithm must decide which subset of potentially thousands of variables actually matters for a prediction, such failures translate directly into bloated models, longer training times, and degraded accuracy.

AIRIME addresses these shortcomings through four coordinated modifications, each targeting a distinct stage of the search. The first, Adaptive k-Average-Best Mutation (AKAB), introduces a mutation mechanism that draws on the average of the k best solutions found so far, with the parameter k adapting over the course of the run. By anchoring mutations to a consensus of strong candidates rather than a single best individual, the strategy reduces the risk of the search collapsing prematurely around one lucky solution. The second enhancement, the Adaptive Searching Process (ASP), dynamically adjusts how the population moves through the search space as iterations progress, shifting the emphasis from broad exploration early on to focused exploitation later, which the authors identify as key to accelerating convergence without sacrificing coverage.

The remaining two mechanisms concentrate on solution quality itself. Enhanced Solution Quality (ESQ) applies refinement steps designed to improve the fitness of individual candidates between generations, effectively polishing intermediate results rather than waiting for the population-level dynamics to do all the work. Gaussian-Based Optimal Solution Mutation (GBOM) then perturbs the current best solution using Gaussian-distributed random variations, a technique that exploits the well-behaved statistical properties of the normal distribution to generate perturbations that are usually small but occasionally large. This gives the algorithm a principled way to escape local optima around the incumbent best solution while still refining it most of the time. Together, the four strategies form an adaptive wrapper around the core rime-ice dynamics, allowing the algorithm to tune its own behavior to the problem at hand rather than relying on fixed parameters chosen by trial and error.

Benchmarking a new metaheuristic credibly requires more than a handful of test functions, and the team put AIRIME through an unusually thorough evaluation. The algorithm was tested on 41 functions drawn from the CEC2017 and CEC2022 test suites, standardized collections of numerical optimization problems designed by the evolutionary computation community specifically to expose weaknesses in search algorithms. These suites include shifted, rotated, and hybrid composition functions that defeat algorithms tuned to simple convex landscapes. AIRIME was compared against a range of classical, recent, and advanced competitors, and the results were subjected to formal statistical scrutiny. Friedman rank tests, which rank algorithms across all test functions and assess whether the rankings differ significantly, and Wilcoxon rank-sum tests, which compare pairwise performance, both validated the algorithm’s superiority and robustness. Sensitivity analyses further examined how each of the four proposed strategies contributed to the overall result, demonstrating the adaptability and efficiency of the individual components rather than attributing success to the ensemble as an undifferentiated whole.

Optimization benchmarks measure performance on continuous mathematical functions, but many of the most consequential applications are discrete. To bridge that gap, the researchers developed a binary version of AIRIME, in which candidate solutions take the form of binary vectors where each bit indicates whether a particular feature should be included or excluded from a predictive model. This binary variant was applied to feature selection tasks across 14 datasets, a scale sufficient to reveal how the algorithm behaves on problems of genuinely different structure and dimensionality. Across these tasks, the binary AIRIME achieved promising results compared with various established binary metaheuristics, the specialized algorithms that practitioners typically reach for when wrapping optimization around classification pipelines. Feature selection matters enormously in practice: removing irrelevant or redundant variables reduces overfitting, cuts computational cost, and often improves the interpretability of the resulting model, a consideration that is especially acute in clinical settings where physicians need to understand why a model produces its predictions.

The most eye-catching demonstration came in the healthcare domain. The researchers embedded AIRIME within a novel Monkeypox disease prediction framework, using the algorithm to select the most informative features from high-dimensional data before feeding them to a classifier. Monkeypox, a viral disease that gained global attention during the 2022 outbreak, presents diagnostic challenges in which rapid, accurate computational triage tools could support clinical decision-making, particularly in settings with limited laboratory capacity. In this framework, AIRIME’s feature selection distilled the input data down to a significantly smaller set of variables while the overall pipeline achieved a classification accuracy of 95.52 percent. The combination of high accuracy and aggressive feature reduction is the crucial point: it suggests the algorithm identified a compact, information-dense signature within the data rather than simply memorizing a large number of weakly informative inputs. For high-dimensional healthcare data, where the ratio of variables to samples is often unfavorable, that efficiency is precisely what determines whether a model generalizes beyond its training set.

The authors frame AIRIME as a significant advancement in big data-driven optimization, combining scalability, adaptability, and robustness to address complex and high-dimensional challenges across domains that extend well beyond healthcare. The design philosophy on display here reflects a broader trend in computational intelligence: rather than inventing entirely new metaphors, researchers are increasingly investing in adaptive mechanisms that let existing algorithms self-tune across problem types. The four AIRIME strategies, mutation anchored to top-k averages, adaptive search dynamics, explicit solution refinement, and Gaussian perturbation of the best solution, are each individually well-motivated, and the sensitivity analyses suggest their benefits are separable and transferable. That modularity could make the enhancements attractive to researchers working with other population-based optimizers.

Caveats remain, as they do with any newly proposed metaheuristic. Benchmark suites, however demanding, are proxies for real problems, and the monkeypox result, while impressive, comes from a single application whose clinical value will depend on prospective validation with independent patient data. The algorithm’s performance across other discrete and mixed-integer problem classes remains to be mapped. Still, the study offers a methodologically careful case: standardized benchmarks, statistical hypothesis testing, sensitivity analysis, and a concrete application demonstrating end-to-end value. Published open access in the Journal of Big Data, the work gives the optimization community both a stronger tool and a template for how adaptive enhancements to nature-inspired algorithms can be evaluated rigorously. If the rime-ice metaphor continues to mature along this trajectory, the frost forming on the surfaces of high-dimensional search spaces may prove to be one of the more productive crystallizations in the field.

Subject of Research: An adaptive improved rime-ice optimization algorithm for feature selection and monkeypox disease prediction

Article Title: AIRIME: a scalable adaptive RIME algorithm for feature optimization and data-driven monkeypox disease prediction

Article References: Elhosseny, M., Abdel-salam, M., Khafaga, D. S., Aldakheel, E. A., Zhong, R., & El-Hasnony, I. M. (2026). AIRIME: a scalable adaptive RIME algorithm for feature optimization and data-driven monkeypox disease prediction. Journal of Big Data. https://doi.org/10.1186/s40537-026-01482-w

Image Credits: AI Generated

DOI: 10.1186/s40537-026-01482-w

Keywords: AIRIME, RIME algorithm, metaheuristics, global optimization, feature selection, big data, machine learning, monkeypox, disease prediction, CEC2017, CEC2022, computational intelligence

At the heart of the research lies a deceptively simple question: how should a robot choose a path that is simultaneously short, smooth enough for its drivetrain to follow, and guaranteed to avoid obstacles? These goals can pull in different directions. The shortest geometric route may hug obstacles so tightly that any sensor noise or wheel slip would cause a collision, while the smoothest route may wander far from the ideal straight line. The team formalized this trade-off in a mathematical path planning model that combines path length, smoothness, and collision avoidance into a single cost function, turning route selection into a constrained optimization problem that an algorithm can attack directly.

The algorithm they chose to modify is particle swarm optimization, a nature-inspired technique modeled on the way bird flocks and fish schools share information. In classical PSO, each candidate solution, called a particle, flies through the search space with a position and a velocity. Every particle is tugged toward the best solution it has personally found and toward the best solution discovered by the whole swarm. The method is celebrated for its speed and simplicity, but it carries a well-known flaw: as iterations accumulate, particles converge on the same region and the swarm loses its exploratory power, a phenomenon researchers call premature convergence. In path planning, that failure mode translates into routes that look good early but miss the globally optimal corridor entirely.

Yang and his colleagues attacked this weakness on two fronts. First, they rebuilt the velocity and position updating equations with dynamic adaptive strategies, meaning the parameters that govern how aggressively particles chase personal versus collective memories change over the course of the run rather than staying fixed. Early in the search, the swarm is encouraged to roam widely across the map; later, the settings tighten so that promising regions are exploited with fine-grained precision. This kind of time-varying tuning is not cosmetic. It changes the balance between exploration and exploitation, the central tension in any stochastic optimization method, and it directly counters the stagnation that plagues the original PSO formulation.

The second innovation is a spiral search strategy, borrowed conceptually from optimization algorithms that mimic spiral motion patterns in nature, which reactivates the algorithm’s global search ability when the swarm shows signs of settling. Instead of letting particles collapse into a single cluster around a local optimum, the spiral mechanism moves them along curved trajectories around promising regions, allowing the swarm to sample the surrounding space systematically rather than randomly. In effect, the algorithm gains a second chance: when convergence stalls, the spiral operator re-expands the search before it contracts again, giving the optimizer multiple opportunities to escape basins of attraction that would otherwise trap it.

To find out whether these modifications actually matter, the researchers designed a battery of four simulation experiments. Two of them are standardized benchmark tests drawn from the CEC2005 and CEC2017 competition suites, internationally recognized collections of numerical optimization functions that are deliberately engineered with deceptive landscapes, many local optima, and shifted or rotated variables. Benchmarking against these suites is the standard way to prove that a new optimizer is genuinely better rather than merely lucky on one toy problem. The remaining experiments place the algorithms inside three obstacle scenes of escalating difficulty: a simple 20 by 20 grid map, an intermediate 30 by 30 grid, and a complex, randomly generated 40 by 40 grid. Crucially, MPSO was not compared against a single rival. It went head to head with eight other heuristic algorithms, including the basic PSO, the whale optimization algorithm (WOA), and an improved hybrid variant the authors call PSO-AWOA.

The benchmark results showed MPSO outperforming all eight competitors on the standardized test functions, establishing its raw optimization credentials before any robotics-specific evaluation. On the simple 20 by 20 grid, MPSO delivered the best overall performance, improving the total cost value by an average of 18.31 percent compared with the basic particle swarm optimization algorithm. That figure matters because the cost function bundles length, smoothness, and safety together, so an 18 percent reduction represents a genuinely better route, not just a marginal tweak in one dimension.

The intermediate map is where the smoothness gains became dramatic. On the 30 by 30 grid, the authors report that MPSO can theoretically improve path smoothness performance by 71.42 percent over PSO-AWOA, an improved whale optimization algorithm, and by 73.91 percent over the standard whale optimization algorithm. Smoothness is not an aesthetic nicety. A jerky path full of sharp turns forces a wheeled robot to decelerate, rotate in place, and accelerate repeatedly, consuming more energy, wearing out actuators, and increasing the time needed to complete a delivery or inspection task. For platforms like Ackermann-steered vehicles, which cannot turn on the spot at all, a smooth, continuously curved trajectory is often the difference between a path that can be followed and one that cannot.

The most demanding test, the randomly generated 40 by 40 grid, simulates the kind of unstructured environment a rescue robot or an autonomous service vehicle might actually face, where obstacle positions cannot be anticipated in advance. Here MPSO succeeded in obtaining the best optimal path trajectory among all the tested algorithms, suggesting that the combination of adaptive parameter tuning and spiral re-exploration scales to problems where the search space is large and riddled with dead ends. The progression from simple to complex maps also serves as a stress test of robustness: an algorithm that wins only on easy terrain is of limited practical value, whereas one that maintains its lead as clutter and randomness increase is far more credible as a candidate for real deployment.

The broader significance of the study lies in what it says about the maturing field of metaheuristic robotics research. Motion planning is formally NP-hard, meaning that no known algorithm can guarantee the optimal path in polynomial time for general environments, which is why the community has long relied on heuristics ranging from genetic algorithms and ant colony optimization to grey wolf optimizers and artificial potential fields. Each generation of methods refines the balance between computation time and solution quality, and this work shows that careful, principled modification of a thirty-year-old algorithm can still yield double-digit percentage gains over recent alternatives. With mobile robots moving from laboratories into warehouses, hospitals, farms, and public spaces, algorithms that consistently produce shorter, smoother, and safer routes translate directly into lower operating costs and higher reliability. The authors, whose work was supported by innovation programs of Shandong Province and Qilu University of Technology, have provided both a competitive new tool and a template for how adaptive strategies and hybrid search operators can be layered onto classical swarm intelligence to keep pace with increasingly demanding robotic missions.

Subject of Research: Mobile robot path planning using a modified particle swarm optimization algorithm

Article Title: Research on efficient mobile robot path planning using a modified particle swarm optimization algorithm

Article References: Yang, Z., Zhao, J., Luo, X., Liu, N., Qiao, J., & Liu, P. (2026). Research on efficient mobile robot path planning using a modified particle swarm optimization algorithm. International Journal of Intelligent Robotics and Applications. https://doi.org/10.1007/s41315-026-00601-3

Image Credits: AI Generated

DOI: 10.1007/s41315-026-00601-3

Keywords: mobile robot, path planning, particle swarm optimization, metaheuristics, obstacle avoidance, path smoothness, whale optimization algorithm, swarm intelligence, optimization algorithms, robotics, CEC benchmarks, collision avoidance

The challenge they tackle is a familiar one in bioinformatics. DNA microarray datasets typically measure the expression levels of thousands of genes, yet they contain samples from only a few dozen to a few hundred patients. In this high-dimensional, low-sample-size regime, most genes are redundant, noisy, or entirely irrelevant to the disease being studied, and feeding all of them into a classifier invites overfitting, inflated computational cost, and unreliable predictions. Feature selection—the process of identifying the small subset of genes that genuinely discriminate between cancerous and healthy samples—is therefore a critical preprocessing step before any classification model is trained.

Feature selection methods come in two broad families. Filter approaches rank every feature according to a mathematical score, such as its statistical association with the class label, and keep the top performers; they are fast but blind to how the chosen features will actually behave inside a classifier. Wrapper approaches, by contrast, evaluate feature subsets by directly measuring the classification accuracy they produce, which usually yields better subsets at a much higher computational price. The new study combines the two in a pipeline: five filter techniques—ANOVA, the chi-square test, Fisher score, Gini index, and information gain—first prune thousands of genes down to the top one hundred, and then a wrapper method refines that pool to the most informative subset.

The novelty lies in the wrapper. WSO is a swarm intelligence algorithm modeled on the foraging behavior of great white sharks. A population of virtual sharks, each representing a candidate solution, moves through the search space governed by equations that mimic three shark behaviors: swimming toward prey detected by wave-like sensing, following scent trails, and converging on the position of the best-performing shark near the target. Velocity updates incorporate a constriction factor to control convergence, while parameters representing the sharks’ smelling and seeing ability govern the balance between exploration of the search space and exploitation of promising regions. The authors built a binary version, BWSO, in which each shark’s position is encoded as a vector of ones and zeros, with a one indicating that the corresponding gene is selected.

The fitness function that guides the search balances two competing goals: classification accuracy and the fraction of features removed. Because diagnostic performance matters most in a medical setting, the accuracy term receives a weight of 0.9, while feature reduction carries the remaining 0.1. The best solutions are passed forward through generations, and the authors report that this solution-based transfer strategy produces faster convergence and better feature-selection performance than a naive search.

To test the framework rigorously, the researchers constructed every possible combination of five filters, three wrappers (WSO alongside the established Ant Colony Optimization and Particle Swarm Optimization), and three classifiers (logistic regression, random forest, and support vector machine), yielding forty-five candidate models. These were evaluated on four public microarray datasets covering colon tumor, central nervous system tumors, breast cancer, and lung cancer. Because no single metric can crown a winner across datasets and classifiers, the team applied a two-stage grading procedure: models were first ranked on each of six performance measures—accuracy, precision, recall, F1 score, G-mean, and Jaccard coefficient—and then re-ranked across datasets to identify the most consistently strong performers. Friedman and Wilcoxon signed-rank tests provided statistical validation of the rankings.

The results were striking. Before any wrapper was applied, filter-based models alone performed poorly, particularly on the colon and breast cancer datasets, where accuracy rarely exceeded 90 percent—a level the authors argue is insufficient for medical diagnosis. Once the wrapper stage was added, one model dominated: GR18, which pairs information gain filtering with BWSO and a support vector machine classifier. GR18 achieved the lowest average grade on three of the four datasets, reached testing accuracies of 0.9951 on the colon dataset and 0.983 on the lung cancer dataset, and consistently recorded the highest area under the ROC curve across all four datasets. It also selected the fewest features of any top-ranked model, confirming that the shark-inspired search prunes redundancy aggressively rather than simply hoarding informative-looking genes.

Comparisons against other swarm-based wrappers reinforced the case. When the same information-gain-plus-SVM pipeline was run with cuckoo search, grey wolf optimization, firefly algorithm, and artificial bee colony wrappers, WSO delivered the best results on the colon and central nervous system datasets and competitive performance on the other two, while converging smoothly within one hundred runs. The authors note that although BWSO’s running time was not always the shortest, classification accuracy—the primary objective in healthcare analytics—was consistently superior. An ablation study fixing the population size at 30 and the accuracy weight at 0.9 showed these settings reliably produced the best trade-off between accuracy and feature reduction.

What makes the work notable is less any single accuracy figure than the demonstration that a relatively young metaheuristic, previously applied to power scheduling, wind forecasting, and cloud scheduling, transfers effectively to genomics. The authors argue that WSO’s explicit balance of exploration and exploitation helps it escape the slow convergence and poor fitness landscapes that plagued earlier feature-selection heuristics. They caution that no single model suits every cancer dataset, which is precisely why their multi-criteria grading framework matters: it offers a principled way to choose among dozens of pipeline combinations rather than relying on one metric.

The researchers see several directions ahead. The same filter-wrapper-classifier architecture could be extended to multiclass medical classification problems and to other high-dimensional domains beyond medicine, and they suggest WSO could serve real-time Internet of Things systems where computational cost is at a premium, or anchor ensemble frameworks that combine multiple feature-selection and classification models. For now, the study offers a vivid example of how borrowing the sensory strategy of an ocean predator can sharpen the search for a molecular needle in a genomic haystack.

Subject of Research: Application of the white shark optimization algorithm to feature selection and classification of high-dimensional DNA microarray cancer data

Article Title: Unveiling the potential of white shark optimization algorithm for high dimensional cancer data classification

Article References: Dash, B., & Dash, R. (2026). Unveiling the potential of white shark optimization algorithm for high dimensional cancer data classification. Discover Artificial Intelligence, 6(1), Article 1386. https://doi.org/10.1007/s44163-026-01808-w

Image Credits: AI Generated

DOI: 10.1007/s44163-026-01808-w

Keywords: white shark optimization, swarm intelligence, feature selection, DNA microarray, cancer classification, bioinformatics, machine learning, wrapper methods, filter methods, high-dimensional data, support vector machine, metaheuristics

Researchers Salwa Khalid Abdulateef, Armaneesa Namaan Hasoon and Israa Rafaa Abdulqader built a framework called DAG–SA that treats ensemble construction as a discrete optimisation problem. From a pool of 420 trained base learners — support-vector machines and linear discriminants paired with feature views drawn from Common Spatial Pattern (CSP), Common Spatio-Spectral Pattern (CSSP) and raw channel log-variance extractors across four frequency bands — the method assembles committees of four to eight members. The committees are structured as small directed acyclic graphs: two internal nodes, each grouping a subset of learners, feed a root node, and every node applies one of five fusion operators, from soft voting and majority voting to a conservative minimum rule and a stacking meta-learner.

The scale of the search space is staggering. A formally specified graph grammar generates roughly 9.7 × 10^19 distinct topologies, ranging from about 2.6 × 10^11 four-member ensembles to nearly 10^20 eight-member ones. Exhaustive enumeration is hopeless, so the team turned to simulated annealing, the classic metaheuristic introduced by Kirkpatrick and colleagues in 1983, which wanders rugged discrete landscapes while its temperature-controlled acceptance of worse moves lets it escape local optima. Each candidate topology is scored on held-out validation data, and the annealer perturbs its current graph through five moves: swapping a leaf, changing a node operator, changing the root operator, adding a leaf, or deleting one.

Crucially, the authors designed the evaluation to be airtight. Spatial filters were fitted only on training splits, support-vector probability calibration used internal cross-validation, stacking meta-learners were fitted on out-of-fold predictions so no node was ever scored on data it had seen, and the test split was read exactly once per method. Every comparison was paired: twenty random seeds controlled the data partition, the search randomness and classifier training simultaneously, so all six methods under a given seed were trained and tested on identical trials. The benchmarks were the calibration recordings of BCI Competition IV Dataset 1, with four subjects, and the nine-subject Dataset 2a, restricted to a binary left- versus right-hand task.

The headline result is a negative one, and the authors report it with unusual candour. On Dataset 1, DAG–SA reached 67.2 percent accuracy and ranked fourth of six methods, behind single-best selection (70.2 percent), a Riemannian tangent-space classifier (69.9 percent) and even random search from the same sampler (69.2 percent). Under a stricter repeated cross-validation protocol, the ordering barely changed: DAG–SA again ranked fourth at 68.2 percent. On Dataset 2a it ranked last, at 64.2 percent, while the compact convolutional network EEGNet dominated at 77.2 percent. The elaborate searched ensembles did not beat simply picking the single best pool member.

Why did a sophisticated search lose to a coin-flip-style baseline? The diagnostic experiments point squarely at the selection signal. Validation accuracy estimated from just 22 to 31 calibration trials correlates only weakly with test accuracy — mean Pearson coefficients of 0.39 on Dataset 1 and 0.28 on Dataset 2a, with unit-level values swinging from −0.66 to 0.87. A single validation trial shifts the score by 3.2 to 4.5 percentage points, so choosing the maximum over thousands of topologies systematically favours candidates whose validation scores are inflated by chance. When the authors gave the annealer budgets up to 10,000 evaluations, its validation score climbed steadily while its test accuracy stayed flat.

The annealing schedule itself also contributed. With an initial temperature of 5 on an objective bounded between 0 and 1, plus a reheating rule that multiplies the temperature by 1.5 whenever progress stalls, 96 percent of proposed moves were accepted at the default budget — the annealer was effectively performing a correlated random walk rather than a targeted descent. Lowering the initial temperature to 0.05 cut acceptance to 40–53 percent and raised validation accuracy to near random-search levels, but test accuracy did not follow. On an enumerable sub-pool of ten learners, where all 78,750 four-member topologies could be scored exactly, both searches reached the true optimum in comparable fractions of runs, confirming that the optimiser was not the bottleneck.

Two further findings sharpen the picture. A same-family rule, which confined each annealing run to either the CSP/CSSP family or the log-variance family, turned out to hurt: half the runs were locked into the less accurate family for each dataset, and relaxing the constraint raised test accuracy by 2.4 and 3.5 points on the two datasets — though even the unconstrained search never surpassed the strongest baselines. And a transfer analysis, rebuilding each subject’s selected topology on every other subject, found only a small and statistically uncertain subject-specific advantage: the own topology won by 2.7 points on Dataset 1, concentrated in a single subject, and by a negligible 0.6 points on Dataset 2a.

The study’s real contribution may be methodological rather than empirical. Every returned ensemble is a compact, fully specified object — named learners, their division between two nodes, three fusion operators — that can be inspected, compared across subjects and rebuilt on new data. Inspecting the outputs is what exposed the in-sample stacking bias and the family confinement in the first place. The grammar, shared sampler and fixed budget make the search auditable: every topology the method can return is counted, and the quality of the space can be measured independently of any search. Perhaps most valuably, the paper offers a transferable lesson for the growing field of search-based neural decoders: the reliability of the selection score, measurable directly as the agreement between selection and held-out accuracy over sampled candidates, bounds what any larger search budget or cleverer optimiser can deliver.

For brain–computer interfaces hoping to reach clinical use, the message is sobering but constructive. Before proposing a bigger search space or a smarter metaheuristic for subject-specific calibration, researchers should first ask whether a short calibration block can rank candidate models at all. The authors suggest variance-reduced selection scores — repeated or bootstrapped cross-validation estimates, log-loss, or penalties on ensemble diversity — as the first target for future work, alongside moves that let the annealer escape its initial feature family and pools enriched with learned representations. In a field where deep decoders are often benchmarked on a single split, a rigorously matched, fully released negative result with a quantified explanation may prove more influential than another incremental accuracy claim.

Subject of Research: Simulated-annealing search over directed-acyclic-graph ensembles for subject-specific motor-imagery EEG decoding

Article Title: Annealed search over directed-acyclic-graph ensembles for subject-specific motor-imagery EEG decoding

Article References: Abdulateef, S. K., Hasoon, A. N., & Abdulqader, I. R. (2026). Annealed search over directed-acyclic-graph ensembles for subject-specific motor-imagery EEG decoding. Machine Learning with Applications, 26, Article 101040. https://doi.org/10.1016/j.mlwa.2026.101040

Image Credits: AI Generated

DOI: 10.1016/j.mlwa.2026.101040

Keywords: brain-computer interface, motor imagery, EEG, simulated annealing, ensemble learning, Common Spatial Patterns, directed acyclic graph, model selection, validation overfitting, BCI Competition IV, subject-specific decoding, metaheuristics