Cooperative autonomous underwater vehicles (AUVs) provide an effective platform for marine environmental monitoring and offshore energy exploration. However, federated learning in constrained underwater networks is challenged by the deep coupling between scarce high-value observations and heterogeneous non-IID sensing data. Existing methods usually ignore the coupling between physical energy consumption and model learning, or aggregate sparse heterogeneous updates uniformly, which may cause inefficient training and pseudo-convergence. To address these challenges, this paper proposes an Energy-Aware Federated Multi-Agent Learning framework, named Energy-Aware FedMARL. The proposed framework formulates cooperative sensing as a constrained Markov decision process and adopts a MATD3-based multi-agent strategy to guide AUVs toward spatially complementary high-value observations under energy constraints. In addition, a quality-aware aggregation mechanism is developed to emphasize sparse local updates containing more high-value samples, with extensions to validation-gain and contribution-aware weighting. Simulation results show that Energy-Aware FedMARL reduces global validation loss by about 30% compared with MAPPO and 68% compared with MADDPG, while improving hotspot discovery and effective federated participation.

Gross, Mansour and Tucker introduced the partial-twuality polynomials for ribbon graphs and investigated the interpolation property of these polynomials. The ribbon group generated by δ and τ acts on set systems as twist ∗ and loop complementation ×, yielding five nontrivial twuality operators: {∗,×,∗× ,×∗ ,∗×∗}. Yan and Jin extended partial-twuality polynomials to set systems, yielding partial-∙ polynomials with ∙∈{∗,×,∗× ,×∗ ,∗×∗}. For partial-∗ polynomials, Zhao and Yan proved that this polynomial is either even, odd, or both even-interpolating and odd-interpolating for every binary delta-matroid. In this paper, we extend this interpolation property to all the remaining nontrivial partial-twualities of binary delta-matroids. Consequently, for every binary delta-matroid and every ∙∈{∗,×,∗× ,×∗ ,∗×∗}, the partial-∙ polynomial is either even, odd, or both even-interpolating and odd-interpolating. We also provide examples to show that the binary assumption is essential.

Edge-side model fine-tuning is increasingly becoming a key technology owing to its inherent advantages in privacy preservation. However, dynamic fluctuations in the resource environments of mobile devices, the conflicts among multiple objective constraints, and the intrinsic resource limitations of the devices collectively pose significant challenges to the intelligence and stability of fine-tuning strategies. To address these challenges, we propose an adaptive fine-tuning strategy based on Proximal Policy Optimization called PPO-AF. This method dynamically selects fine-tuning strategies based on the current device state, achieving comprehensive optimization of model performance, resource consumption, and user experience. Experimental results show that PPO-AF significantly outperforms existing methods in terms of energy efficiency and stability. Specifically, it improves the reward on unit resources by 69.0% and 49.6% compared to A2C and DQN, respectively. In terms of training efficiency, it is 44.4% faster than A2C and 52.4% faster than DQN. Furthermore, it achieves 37.5% higher episodic rewards than recent QR-DQN.

In response to the common problems of serious frost heave damage such as uplift, dislocation, and collapse in the slope protection of earth-rock dams in cold regions, this paper established a model for moisture migration during soil freezing by using the soil-water potential gradient and permeability coefficient within the frozen boundary. It also proposed a calculation method of ice loads considering the characteristics of the slope protection of earth-rock dams. As a result, a thermo-hydro-mechanical coupling model for frost heave in earth-rock dam slope protection was developed, which took into account both moisture migration and ice loads. Taking the slope protection of a certain earth-rock dam in a cold region as the research object, this study investigated the variation patterns of the temperature field, hydraulic field, and displacement field, as well as the frost heave damage mechanism of the earth-rock dam slope protection. The results indicated that the frost heave damage to slope protection was mainly caused by the combined effects of the sand gravel cushion and the dam filling ‘s own frost heave, moisture migration, and ice loads. The effects of moisture migration and ice loads on the slope protection's frost heave damage were significant. Moisture migration inhibited the movement of the frozen front from sand gravel cushion into the dam body. Moisture from the unfrozen areas of the dam continued to migrate toward the frozen front, prolonging the ice-water phase change process. This increased the ice volume content in the frozen area, thereby intensifying the frost heave damage to the slope protection. Under the frost heave of the sand gravel cushion and dam filling itself, cracks and uplift occurred on the surface of the concrete slab and at the joints. Then, the ice loads led to uneven deformation, dislocation and void of the concrete slab. After the frost heave process, the effects of the reservoir water infiltration and wave scouring resulted in the sand gravel cushion loss. Through multiple freeze-thaw cycles, this eventually caused the failure of the earth-rock dam slope protection. The damage at first operating period to the earth-rock dam slope protection was mainly due to frost heave, while damage at long-term operating period resulted from the combined effects of freeze-thaw and frost heave. The results provide scientific evidence and technical support for the design and operation management of earth-rock dam slope protection in cold regions and fill the gaps of the slope protection design in design code for rolled earth-rock fill dams (Chinese standard).

In cold-region steep rock slopes, ice-wedge freeze-thaw action imposes cumulative damage on the long-term stability of toppling perilous rocks. However, efficient and interpretable methods for predicting the intrinsic coupling among geometric evolution, mechanical response, and instability thresholds are still lacking. Focusing on the progressive instability of toppling perilous rocks governed by ice-wedge freeze-thaw processes, this study develops an analytical expression for the stability evolution of such rocks under multiple freeze-thaw cycles based on the water-ice phase transition mechanism, and introduces a physics-informed neural network (PINN) to establish a unified prediction framework integrating ice-wedge freeze-thaw action, structural-plane evolution, and stability response. The results indicate that the instability of toppling perilous rocks in cold regions is not controlled solely by rock mass strength degradation, but rather represents a geometry-dominated progressive-catastrophic failure process governed by irreversible evolution of controlling structural planes. Ice-wedge freeze-thaw action promotes directional and organized fracture propagation along structurally dominant directions that control toppling instability, gradually forming structural failure zones that govern the transition from localized damage to global instability. This process is further characterized by a pronounced cumulative amplification effect associated with repeated ice-wedge splitting. The proposed PINN framework maintains physical consistency while efficiently capturing the continuous evolution of rock stability under ice-wedge freeze-thaw conditions and substantially reducing computational cost. Through the deep integration of analytical ice-wedge freeze-thaw mechanics with PINN, this study elucidates the positive feedback mechanism between organized fracture development and toppling instability, providing an efficient and physically consistent new approach for long-term stability assessment and risk early warning of perilous rocks in cold regions.

Video is a dominant traffic type in mobile networks. D2D communication, one of the most promising technologies for 6G, enables efficient localized distribution of content and thus reduces latency and bandwidth consumption for content delivery. Discovering potential mobile user devices and caching promotional content or short video content in D2D self-organizing communication-based social networks requires overcoming challenges, such as heterogeneous structural features of user interactions and user state changes. We propose a graph neural network based graph short video interaction aggregation (GSVIA) framework that integrates user information from the perspective of user interaction behavior and social features, and also propose an InfTMC algorithm for maximizing the weights of a set of caching nodes using aggregate function submodularity. Experimental results under two different weighted independent cascade (IC) models show that the InfTMC algorithm improves the influence spread by an average of 46.80% and 45.44% compared to the existing algorithms.

Computing power measurement faces the challenges of heterogeneity of devices, poor scalability with workloads, and discrepancies between measurement methods and models. In this article, we propose a computing power measurement framework called AdaptPerf to address these issues. AdaptPerf uses the technique of neural architecture search to build a hypernetwork of benchmark models for heterogeneous devices and adaptively selects the ones suitable for them. It achieves equivalency in computing power measurement through kernel-level latency analysis. We further prove that the adaptive benchmark measurement model selection problem (ABMSP) is an NP-hard problem and introduced a deep reinforcement learning approach to solve it. The theoretical analyzes show that the proposed method can guarantee that the obtained solution approximates the Pareto frontier. We implement AdaptPerf in a real-world environment, constructing a hypernetwork of over $4.2 \times 10<^>{20}$ models for heterogeneous devices to autonomously select the benchmark model adapted to the devices. We use different tools, such as Nsight Compute, Nsight System, to analyze and evaluate the feasibility and performance of AdaptPerf. It is found that the models generated by AdaptPerf increase computational throughput by 46.63% and improve data reuse rate by 93.7%, compared with more than 60 traditional models. AdaptPerf also exhibits superior load adaptability and model complexity compared with existing methods, such as MLPerf.

In Hierarchical Split Federated Learning, the efficiency and quality of model training face challenges due to the varying data quality across nodes and the dynamic changes in their computational and communication capabilities over time. To address this issue, this paper investigates the node selection problem for minimizing system delay under model anomaly constraints and proposes an online node selection algorithm based on Contextual Combinatorial Multi-Armed Bandit (CS-MAB). The algorithm evaluates node data quality through model anomalies and dynamically selects nodes based on their computational and communication capabilities to optimize training delay and improve model quality. Compared to existing algorithms like CS-UCB and FedAvg, CS-MAB demonstrates significant advantages in reducing delay and improving model accuracy. Experimental results show that the proposed algorithm reduces delay by 53.72% and 49.45% on MNIST, FashionMNIST, and CIFAR-10 datasets, respectively, and achieves up to a 41.49% improvement in accuracy.

Due to the inherent insensitivity to salt concentration, membrane distillation (MD) has emerged as a promising technology for high-salinity wastewater treatment. However, scaling remains a critical challenge hindering the widespread application of MD. This study proposes a solution through the development of triblock polystyreneblock-polydimethylsiloxane-block-polystyrene (SDS) membranes with precisely engineered pore structures that effectively address this limitation. The results demonstrate that controlled pore size reduction, which was achieved via selective swelling fabrication, significantly mitigates both surface scaling and intra-pore crystallization. The SDS membrane shows remarkable performance stability under diverse high-salinity conditions, including those high-salinity solutions containing common inorganic contaminants such as CaSO4. Furthermore, when implemented in membrane distillation-crystallization (MDC), the SDS membrane maintains stable performance under a continuous operation for 168 h with saturated NaCl feed, successfully producing high-purity water while simultaneously recovering salt crystals. These results demonstrate that controlled pore size reduction represents an effective strategy to mitigate membrane scaling while maintaining performance stability, offering significant advantages for industrial implementation in zero liquid discharge systems treating high-salinity wastewater.

To address the issues of high energy consumption and slow convergence speed in Split learning, we propose a novel framework Batched Split Learning (BSL). We group edge nodes and deploy different device-side models in each batch, then deploy the corresponding server-side model on the edge server. Edge nodes utilize the technology Over-the-Air Computing to transmit crushed data to servers for server-side model training, and feed the gradient of the crushed data back to the nodes to perform model updating. For each group, after a certain number of training rounds unique device-side models are generated, which can be shared across different groups and finally aggregated to obtain a global model. To optimize the training latency and transmission energy cost of BSL, we define a mixed integer nonlinear programming problem. We use the branch-and bound method to solve this problem and also design a heuristic algorithm for it. We take thorough experiments to evaluate the performance of proposed methods. It can be found that the latency of our proposed BSL framework can be about 78% less than Vanilla Split Learning. Compared with mainstream Parallel Split Learning scheme, our energy consumption is about 20% less. Furthermore, our proposed Branch-and-Bound based algorithm reduces the energy consumption by 55% over the energy-efficient clustering method using random update (EECRU) algorithm and 47% over the greedy energy efficient clustering scheme (GEECS) algorithm. Moreover, our proposed Multi-objective Heuristic algorithm reduces 37.81% and 26.18% than EECRU and GEECS algorithms respectively.

The potential privacy breaches in centralized artificial intelligence model training have raised significant public concern. Hierarchical federated learning, as a technology addressing privacy and network efficiency issues, coordinates local devices using edge servers for model training and parameter updates, thereby reducing communication with central cloud servers and diminishing the risk of privacy leaks. However, in this context, the rise of node selfishness presents a significant challenge, undermining training efficiency and the quality of local models, thereby impacting the overall system's performance. This paper addresses the issue by introducing a virtual node selfish queue to characterize dynamic selfishness, considering both training costs and rewards, and formulating the problem of maximizing model quality within the bounds of controlled node selfishness. Utilizing Lyapunov optimization, this issue is divided into two subproblems: controlling the quantity of node data and optimizing node associations. To solve these, we propose the Data Quantity Control and Client Association (DCCA) algorithm, based on the Hungarian method. This algorithm is shown to ensure boundedness, stability, and optimality in the system. Experimental results demonstrate that the DCCA algorithm enhances model quality by 8.43% and 13.83% compared to the Fmore and FedAvg algorithms, respectively.

In traditional model training, data from certain nodes are reused, which not only wastes computational resources but also may lead to insufficient model generalization capability. This chapter addresses this problem by optimizing the quality of SL models without increasing the training latency by proposing a node selection method that requires initial nodes not to reuse local data and takes over the training task by introducing new nodes. Defining this problem as a quality optimization problem, it can be shown that it is an NP-hard problem. To ensure that the training delay does not exceed the initial limit, the training time of the slowest training node among the original nodes is used as a delay constraint. To solve this problem, the model quality optimization problem under the delay constraint is proposed and an approximate node selection algorithm ANS is designed to select the node that satisfies the delay constraint and maximizes the amount of data. In order to demonstrate that the proposed node selection scheme can not only effectively optimize the quality of the BSL model, but also achieve better data volume maximization while ensuring the delay constraint. The designed ANS algorithm is compared with RTNS and GCNS algorithms, and the experimental results show that the approximate node selection ANS algorithm improves the model quality by 25% over the RTNS algorithm and 15% over the GCNS algorithm.

Selective swelling of block copolymers as an emerging process to prepare ultrafiltration membranes is receiving growing interests. Herein, we report that very little dosages of carbon nanotubes (CNTs) are able to significantly enhance both the separation and mechanical performances of melt-spun polysulfone-block-poly(ethylene glycol) (PSF-b-PEG) hollow-fiber membranes. CNTs are adequately dispersed in the block copolymer by melt processing, and exhibit pi-pi interaction to the PSF continuous phase but repulsion to the PEG dispersed phase. The incompatibility between CNTs and PEG leads to interfacial gaps between CNTs and the PEG phase, thus providing another set of pores facilitating water permeance. Both dosages and aspects of CNTs significantly influence the pore structure and performances of the membranes. Higher dosages of CNTs produce more interfacial gaps and lead to increased porosity and permeance. While CNTs with lower aspects tend to be distributed in the PSF phase, thus producing smaller pores and decreasing permeance by refraining selective swelling to a larger degree. The hollow-fiber membrane doped with 0.01 wt% CNTs having a diameter of similar to 10-20 nm and a length of similar to 50 mu m shows a water permeance increased by three times and a rejection increased by 1.6 times. Moreover, thus-doped membrane exhibits over 1.5 times increase both in tensile stress and the strain at break and multiple times increase in swing tolerance. Such an extremely low dosage of CNTs synchronously boosting membrane permeance, rejection, and mechanical properties is highly desired in practical applications and is expected to be extended in the performance-upgrading of other membranes with multiphases.

In Hierarchical Federated Learning, Split Learning is introduced to support cooperative training among resource constrained devices and edge servers. While high delay cost is a critical issue that needs to be addressed during model training, excessive energy consumption poses a fundamental threat to the system. In this paper, we investigate the problem of minimizing the total delay cost under the constraint of the energy cost for cooperation between nodes. We propose an online cooperative task allocation algorithm OTNA based on Lyapunov optimization theory, which can dynamically changing capabilities and resource states of heterogeneous devices, and coordinate them efficiently to improve device-side model training performance. We prove that there is an upper bound of OTNA through theoretical analysis. We also evaluate the performance of the algorithm through several experiments. Compared with SFL and RNCA, it is found that OTNA can save delay cost up to 18.83

With the rapid development of artificial intelligence, big data, and distributed computing technologies, hierarchical federated learning has emerged as a widely studied distributed machine learning framework. In hierarchical federated learning, edge servers are deployed between cloud servers and mobile devices, efficiently receiving local models from nearby mobile devices and performing edge model aggregation. Node collaboration in hierarchical federated learning can reduce training costs and improve model quality while protecting data privacy. However, data security risks and resource consumption during model training can reduce the willingness of mobile devices to participate. Additionally, collaborative nodes are often heterogeneous, facing issues such as skewed datasets and imbalanced capabilities. Therefore, this paper proposes a deep reinforcement learning-based incentive mechanism for node collaboration, aimed at maximizing node benefits. A node collaboration strategy optimization model is then constructed using the Markov decision process framework, and the NCIA algorithm, based on deep reinforcement learning networks, is designed. Finally, through extensive simulation experiments, the proposed NCIA algorithm is demonstrated to improve model accuracy by 5.28% and 14.22% compared with the CCEG and FedAvg algorithms, respectively.

The effort level of edge nodes to perform outward dissemination for short video collaborative caching system in mobile edge is private information. Meanwhile, due to the limited bandwidth used by edge nodes for short video caching data dissemination, the nodes would like to put less dissemination effort to get higher revenue. To address the issue of selfishness in edge collaborative caching systems for short video applications under incomplete information constraints, we design an incentive mechanism based on contract theory. Our proposed mechanism supports pricing for targeted caching services among different edge caching nodes to achieve trade-off between user access requests and caching service supply. We investigate the optimal contract design problem under the budget constrains that short video application service providers can provide. Based on this, the OCSA incentive mechanism is designed with two steps under budget constraints. First, we give the analytic solution without budget constraint. We simplify the problem using the deterministic equivalent method and obtained the optimal solution under the assumption of sufficient costs. Secondly, we transform the contract design problem under cost constraints into a 0-1 knapsack problem and design the contract allocation algorithm for the short video mobile edge collaborative caching system to solve it. The performance of the proposed OCSA is evaluated and it is found that the caching content propagation range of the short video mobile edge caching collaborative system is 20.4

In order to investigate the freezing damage problem of berms of earth-rock dams in cold regions, an earth-rock dam in a cold region was selected as the research object in this study. A finite element model, considering the effect of thermo-hydro-mechanical coupling, has been developed to solve the problem by combining the characteristics of the earth-rock dam. The whole process of freezing damage of berms under the influence of the reservoir level and the water migration of the dam filling was investigated, and the laws in temperature, humidity and displacement of earth-rock dams were analyzed. The calculated displacement field was then compared with the measured frozen deformation data to validate the results of the finite element simulation. The results showed that the freezing influence range of the dam slope was about 2 m, the range of temperature influence on the dam slope mainly depended on the depth of freezing, and the temperature change in the shallow range (0-2 m) of the dam slope was influenced by the outside air temperature. Also, the internal temperature of the dam body was small relative to the shallow dam slope, and there was a certain hysteresis. In addition, the effect of negative temperature was such that the shallow pore water phase of the dam slope turned into ice, manifesting macroscopically as a reduction in unfrozen water content. Water phase change, water migration from the dam filling to the dam slope, and the movement of the ice peak towards the dam body were found to be the main causes of berm freezing and expansion damage. The calculated amount expansion (due to freezing) of the dam slope was found to be in the range of 20-30 cm with a maximum value of 36 cm, which was consistent with the measured results. It was also found that the freezing and expansion damage of the berm is mainly caused by the joint action of freezing and expansion of the soil and rock mixture such as gravel bedding and dam filling, as well as the ice thrust force. It is expected that the results of this research can provide a basis for the design of berms of earth-rock dams in cold regions.

Hemodialysis has been used as the primary treatment for patients with end-stage renal disease, however, the current hemodialysis membranes still need complicated modifications to enhance the hemocompatibility and overcome the additive leaching issue. Herein, we prepare hemodialysis hollow-fiber membranes (HFMs) via the melt spinning and selective swelling of the block copolymers of polysulfone (PSF) and polyethylene glycol (PEG), PSF-b-PEG. PSF-b-PEG is first melt-extruded to form dense hollow fibers, and then soaked in selective solvents to transform the PEG microdomains into nanopores following the mechanism of selective swelling-induced pore generation, thus producing HFMs with three dimensionally interconnected porosities. As neither additives nor involatile solvents are involved in the manufacturing process of the HFMs, no elution of any organic matters could be detected for the HFMs during hemodialysis. The HFMs possess a symmetrical structure ensuring tight selectivity, and hydrophilic PEG chains are enriched on the pore surfaces, thus endowing the membranes with enhanced hydrophilicity and durable biocompatibility. We systematically investigate the effect of PEG contents and swelling conditions on pore sizes, porosities, surface properties, and consequently the hemodialysis performance. The optimized HFMs reject >99 % serum albumin while clear similar to 70 % middle molecular toxins such as lysozyme and 93-95 % small molecular toxins including urea, phosphate and creatinine. This work provides a strategy to prepare elution-free hemodialysis membranes by taking advantage of selective swelling of block copolymers to balance high protein retaining and high clearance of middle and small molecular toxins, and demonstrates their superiority in hemodialysis performance and safety than conventional membranes.

To optimize short video data dissemination in edge networks, we propose a novel 3-layered network model. Based on the model, we investigate the optimal edge caching node selection problem, which is proved NP-hard. We also analyzed the upper and lower bounds of the data dissemination range in our proposed model. We developed the Graph-Attention-Based Cache Propagation for Degree Calculation (GACPD) to predict the dissemination scale of short video data for each caching node, and the Graph Embedding and GAT-based Caching Node Selection (GEG) algorithm to select the optimal caching nodes. We implement the GEG algorithm and evaluate its performance on both real and simulated datasets. It is found that GEG can reduce the backbone network traffic by 30% to 50%, as well as with 18% to 30% network bandwidth utilization improvement, compared with the existing ICS and CDA algorithms.

The strength deterioration of soil-rock mixtures (SRM) subjected to freeze-thaw (F-T) cycles leads to instability and failure of upper engineering structures in cold regions. However, the mutual feedback response mechanism pertaining to the changes of pore and strength in SRM under F-T cycles are rarely addressed. Nuclear magnetic resonance and triaxial tests were carried out to study the pore structure characteristics and strength response patterns of samples. A correlation model of SRM porosity and strength deterioration was first proposed under F-T cycles, and the model rationality was verified by test data. The results demonstrated that the pore connectivity and porosity increased throughout the F-T process, with the T2 spectral distribution curves exhibiting three peaks. Among these peaks, the main peaks underwent slight changes, while the secondary and micro peaks presented significant changes. Before 3 F-T cycles, the pore distribution evolved to small pores uniformly, followed with the large pores increasing and the micropores disappearing. With increasing of F-T times, the strength and cohesion of SRM experienced a drastic decline, while the internal friction angle demonstrated a slight decrease accompanied by fluctuations. Based on the analysis of test results, a correlation model regarding the porosity and strength deterioration was proposed through the relationship between the micro-structure evolution and the macro-mechanical response during F-T cycles. Furthermore, intrinsic mechanism of SRM strength deterioration under F-T cycles was revealed by considering the pore structure characteristics. The results can provide theoretical insights for the analysis of F-T disaster mechanism and prevention of SRM in cold regions.