.
Federated Learning (FL) enables privacy-preserving collaborative training across decentralized clients, yet its efficacy is severely challenged by heterogeneous local data distributions. While this heterogeneity is well known to cause unstable optimization and degraded generalization, it also gives rise to a deeper, underexplored tension between the generic utility of the global model and the personalized needs of clients. We first systematically investigate this tension via a dual-criterion evaluation protocol, revealing that a single shared global model is often insufficient to excel at both generic and personalized objectives simultaneously, and that conventional single-metric evaluation overlooks critical client-side utility. Motivated by this finding, we propose FedDHP, a novel framework that coordinates a shared backbone and role-specialized dual classifiers to jointly support global generalization and personalized adaptation. FedDHP employs asymmetric data augmentations for the generic classifier to learn robust discriminative features, and the personalized classifier to fit local distributions from weakly augmented views. Through adaptive knowledge distillation with logit adjustment and representation alignment, FedDHP reliably transfers strong local performance into generic gains for the global model under heterogeneous label distributions, while a post-SVD low-rank compression mechanism minimizes communication overhead. Extensive experiments across multiple datasets and heterogeneous settings demonstrate that FedDHP consistently achieves superior global and personalized accuracy, significantly reducing communication cost relative to state-of-the-art baselines.
At present, low-dimensional chaotic systems have deficiencies such as insufficient randomness, limited key space, and uneven sequence distribution, which make the security performance and anti-attack ability of the color image encryption algorithms constructed based on them insufficient and easy to be cracked, thus restricting the application of chaotic theory in the field of image encryption. In response to the above issues, this paper first constructs a three-dimensional dynamical system and proposes a three-dimensional logarithmic-exponential type trigonometrically (sin, cos) coupled chaotic dynamical system (3D-LECCS), this 3D-LECCS is explicitly defined as a globally bounded volume-expanding hyperchaotic system. Meanwhile, through bifurcation diagrams, Lyapunov exponents, sample entropy, permutation entropy, 0–1 chaotic state determination, NIST, and multistability analysis, the hyperchaotic characteristics and complex phase space motion behavior of 3D-LECCS are verified. Secondly, by leveraging the cryptographic properties of this system, a dynamic mapping cascaded encoding encryption mechanism for color images driven by 3D-LECCS (LECCS-CIEA) is designed. This scheme builds a strong nonlinear coupling correlation among the RGB three channels of color images through dynamic spatial partitioning cross-channel remapping and encoding cascaded XOR diffusion strategy. Finally, through theoretical analysis and simulation experiments, it is verified that LECCS-CIEA has good security and anti-attack ability. At the same time, the excellent chaotic performance of 3D-LECCS is also verified, confirming its reliability as an encryption entropy source.
Vibrio anguillarum (V. anguillarum) causes vibriosis in aquaculture, with pathogenicity regulated by quorum sensing (QS). Inhibiting QS is a promising anti-virulence strategy. Plant-derived compounds are attractive due to their safety and low resistance potential. Here, we screened 33 natural compounds and found that esculetin, a coumarin, inhibited V. anguillarum with a minimum inhibitory concentration (MIC) of 40 mg/L. At sub-inhibitory concentrations, esculetin effectively reduced biofilm formation and its key component extracellular polymeric substances (EPS) by 45.84
The gut bacterial community plays a vital role in adjusting host adaptation to temperature variation, a key factor influencing physiology. Currently, the effects of temperature on the bacterial community of sea urchins remain unknown. In this study, the structural alterations and community reassembly of the gut bacterial community in sea urchins (Strongylocentrotus intermedius) under different temperatures (13 degrees C, 16 degrees C, 19 degrees C, 22 degrees C, and 25 degrees C) were investigated by high-throughput sequencing. After 30 days of exposure, high temperatures increase bacterial diversity and alter the structure of the bacterial community. Proteobacteria, Bacteroidota, and Firmicutes were the dominant bacterial communities of S. intermedius, accounting for more than 80% of the total relative abundance. At 13 degrees C, 16 degrees C, and 19 degrees C, Desulforhopalus, Cohaesibacter, and Pseudomonas were the dominant genera, with relative abundances of 34.25%, 11.71%, and 31.49%, respectively, whereas Ruegeria (30.21%) and Sphingomonas (33.23%) became dominant at 22 degrees C and 25 degrees C. Functional prediction based on the KEGG database indicated that key metabolic pathways, including amino acid, carbohydrate, energy, and lipid metabolism, exhibited significant alterations across the different temperatures (P < 0.05). Co-occurrence network analysis revealed that gut bacterial community network complexity was higher at 13 degrees C and 25 degrees C than at other temperatures, with the 13 degrees C and 25 degrees C groups exhibiting the greatest number of nodes (99) and edges (824). In addition, temperature fluctuations strengthened the deterministic assembly of the gut bacterial community, increased the host's selective effect on gut bacteria, and reduced stochasticity. This study aims to provide some theoretical basis for the temperature effects of the gut bacterial community in the sea urchin.
In complex ground environments, conventional RRT* often suffers from poor path quality and slow expansion during robot path planning. To address these issues, this paper proposes GEAR-RRT* (Goal-guided, adaptive informed-Ellipse sampling, layered obstacle-Avoidance expansion, and cost-driven Rewiring), which constructs a collaborative optimization mechanism across the three stages of sampling, expansion, and rewiring. First, the proposed method employs an adaptive informed ellipse to concentrate sampling within feasible regions while dynamically adjusting the informed-ellipse sampling domain, and further integrates Halton-directional hybrid sampling to generate high-quality candidate samples within that domain. Meanwhile, a layered expansion strategy is adopted: the planner first performs direct goal connection for rapid progress toward the goal; when this expansion is blocked by obstacles, it switches to local multi-directional offset to search for feasible expansion directions; if this still fails, an adaptive Artificial Potential Field is introduced to guide subsequent expansions until a feasible path is found. Next, a multi-factor rewiring parent selection strategy is used to optimize path length, safety clearance, and turning angle, while cubic B-spline smoothing is applied to improve path continuity. Finally, GEAR-RRT* is evaluated in five simulation environments as well as in joint ROS and physical-robot validation and is compared with five improved RRT* variants. The results demonstrate that the proposed method achieves superior overall performance in planning time, path length, and safety clearance.
