Machine learning (ML) is highly effective for accurate encrypted malicious traffic identification by using high-quality training data. In fact, obtaining such data is costly and challenging. As a result, many ML-based models are inevitably trained on low-quality data and perform poorly. To enhance performance, some methods utilize various sample selection techniques to choose confident samples for model training. However, they often rely on a single metric for this selection, which restricts their adaptability across diverse datasets and noise conditions. In this paper, we propose a robust framework BAPTISM for identifying encrypted malicious traffic with low-quality training data. Particularly, BAPTISM selects a suitable base model for each task, and trains it with early stopping to generate traffic representation before overfitting occurs. Then, we devise an adaptive metric selection strategy to select confident samples. By employing two metrics (JSD and CSD) to assess the characteristic of traffic representation from distinct perspectives, we find the more proper metric for each class and apply it for confident sample selection. According to the confident samples and selected metric for each class, we develop a label correction tactic which adapts to class nature to improve the quality of training data. Finally, we employ parallel training strategy to train the base model with the corrected data, further mitigating the impact of low-quality data. We conduct experiments across three real-world malicious traffic datasets with various noise settings. The results demonstrate that BAPTISM is compatible with different base models and outperforms across noise ratios ranging from 20% to 90%. Meanwhile, BAPTISM consistently selects the confident samples with the highest purity and volume under each setting.

The multivariate multifractal spectrum (MV-MFS) theory characterizes joint multifractal properties in multivariate signals. However, existing MV-MFS estimation algorithms, such as those based on wavelet leaders and probabilistic measures, suffer from low accuracy for short sequences and are sensitive to data trends. To address these issues, we propose a detrended fluctuation analysis-based MV-MFS (DFA-MVMFS) estimation algorithm. This involves constructing a bivariate joint root mean square (BJ-rms) function, deriving a bivariate joint scaling function from the partial derivatives of the BJ-rms function, obtaining the bivariate multifractal surface spectrum via multivariate Legendre transform, and extending to themultivariate case. Experimental validation using binomialmultiplicative cascades (BMCs) and IPIX radar data demonstrates the superiority of the proposed DFA-MVMFS over benchmarks, and further classification experiments based on the spectrum features using a ResNet network confirm its effectiveness; notably, on 1024-point short sequences, DFA-MVMFS achieves approximately 14% higher detection accuracy compared to wavelet-leaders-based MFS (WL-MVMFS). Furthermore, DFA-MVMFS analysis highlights its potential for target detection in IPIX radar signals.

X-ray emission spectroscopy (XES) and optical Thomson scattering (OTS) are two of the most indispensable diagnostics for characterization of plasma parameters such as Te and ne in intense laser-produced plasmas (LPPs), but both can also provide weighted-average values owing to the line-of-sight, space, and time-integrated effects of optical/X-ray diagnostic techniques, which has not been thoroughly investigated. In this work, for the first time, the consistency and discrepancy between Te from XES and OTS are examined using intense lasers with a large focal spot to irradiate the smaller tip of a cylindrically symmetric titanium wire. Experimental results reveal that Te profiles derived from the two methods are approximately consistent in the earlier period of 1 ns, but exhibit different trends with increasing time. In addition, the impacts of the three integration effects on the interpretation of Te are assessed through an analysis of three types of XES methods, namely, time- and space-resolved, time-resolved but space-integrated, and time-integrated but space-resolved. Our findings indicate that the line-of-sight integration effect causes uncertainties of less than 5% in the electron temperature of corona LPPs, time-integrated XES can replace time-resolved measurements with an uncertainty of 10% for 1–2 ns laser duration, and space-integrated XES reflects the high-temperature region within 200 μm in front of the target surface, ∼10% lower than the maximum. These insights establish a foundation for the applications of experimental data in LPPs, particularly in the fields of laser-driven inertial confinement fusion and high-energy-density physics.

Low-density non–local-thermodynamic-equilibrium plasmas in intense radiation fields occur widely in inertial confinement fusion and astrophysics. Understanding the X-ray spectrum and the atomic kinetics of such plasmas is therefore of great importance. However, the creation of uniform-density nonequilibrium plasmas in intense radiation fields in the laboratory and the measurement of their spectra with high resolution are challenging tasks. Here, we present a new method to produce such a uniform aluminum plasma and explore photon-induced kinetics and relevant atomic physics by measuring its spectrum. It is observed that in the presence of an external radiation field, the satellites q, r and a–d of the He-α resonance line are greatly enhanced compared with the satellites j, k, l. Analysis of atomic kinetics reveals that this effect of intense radiation is due to competition between the photoexcitation and autoionization processes. With this effect taken into account, simulated spectra are able to reproduce the measured spectra quite well.

The increasing integration of renewable energy sources into smart grids presents substantial challenges in solving the nonlinear and nonconvex optimal power flow (OPF) problem. This paper proposes a comprehensive OPF model that incorporates conventional thermal generators, solar photovoltaic generators, and hydroelectric power generators, while effectively addressing the uncertainties associated with renewable power generation. A lognormal probability distribution models solar irradiance variability in solar generators, while a Gumbel distribution captures water availability fluctuations in hydro generators. The paper proposes a novel hybrid optimization approach, hybrid osprey-salp swarm optimization (HOSSO), to solve this complex OPF problem. The HOSSO leverages the exploration-exploitation balance of Osprey Optimization alongside the adaptive leadership and follower dynamics of Salp Swarm Optimization. The proposed methodology is validated on Institute of Electrical and Electronics Engineers 30-, 57-, and 118-bus test systems across five distinct optimization scenarios: economic cost minimization, emission cost minimization, combined economic-environmental cost minimization, voltage deviation penalty cost minimization, and renewable generation uncertainty penalty cost minimization. The model incorporates reserve and penalty costs for renewable generation uncertainty and integrates carbon emission taxation to enhance grid reliability and sustainability. Comparative analysis against classical and hybrid optimization techniques demonstrates the superior performance of HOSSO across most test scenarios, consistently achieving competitive solutions while satisfying system constraints and stability requirements. The algorithm delivers improvements ranging from 0.4% to 17% in cost minimization, 3%-23% in voltage deviation minimization, and 2%-8% in uncertainty management over competing methods, with performance advantages becoming increasingly pronounced as system scale grows. The algorithm exhibits rapid convergence within 20-50 iterations, effectively avoids local optima, and proves well-suited for both single- and multiobjective OPF problems in renewable energy-integrated power systems. The results highlight HOSSO's potential for real-time power system applications and its adaptability to smart grids with high renewable energy penetration.

Safe and efficient autonomous driving in dense traffic is fundamentally a decentralized multi-agent coordination problem, where interactions at conflict points such as merging and weaving must be resolved reliably under partial observability. With only local and incomplete cues, interaction patterns can change rapidly, often causing unstable behaviors such as oscillatory yielding or unsafe commitments. Existing multi-agent reinforcement learning (MARL) approaches either adopt synchronous decision-making, which exacerbate non-stationarity, or depend on centralized sequencing mechanisms that scale poorly as traffic density increases. To address these limitations, we propose Topology-conditioned Stackelberg Coordination (TSC), a learning framework for decentralized interactive driving under communication-free execution, which extracts a time-varying directed priority graph from braid-inspired weaving relations between trajectories, thereby defining local leader-follower dependencies without constructing a global order of play. Conditioned on this graph, TSC endogenously factorizes dense interactions into graph-local Stackelberg subgames and, under centralized training and decentralized execution (CTDE), learns a sequential coordination policy that anticipates leaders via action prediction and trains followers through action-conditioned value learning to approximate local best responses, improving training stability and safety in dense traffic. Experiments across four dense traffic scenarios show that TSC achieves superior performance over representative MARL baselines across key metrics, most notably reducing collisions while maintaining competitive traffic efficiency and control smoothness.

Modern world demands the high technology which can solve the current issues and future problems. In present scenario an electric power driven dual-mode electric bicycle will help to solve the major problems of fuel prices, especially the petrol which is rising steadily day by day. A novel solution to these problems in the form of dual mode e-bicycle with an integrated IoT based smart locking system is presented in this work to address the evolving needs of urban commuters. The e-bicycle’s dual-mode functionality allows riders to seamlessly transition between manual and electric power assistance, catering to varying terrains, distances, and user preferences. Furthermore, the integrated smart locking system, controlled via a user-friendly mobile application, provides enhanced security features, including remote locking and unlocking. Key objectives include enhancing personal mobility, optimizing power distribution, and improving battery technology. The IoT-based smart locking system ensures security and convenience, while design considerations focus on ergonomics and safety. This research demonstrates the potential of combining innovative technologies with sustainable transportation options to address the growing challenges of urbanization and promote eco-conscious mobility solutions.

Zero-shot Composed Image Retrieval (ZS-CIR) involves diverse tasks with varied visual manipulation intents across domains, scenes, objects, and attributes. A key challenge is that existing datasets contain limited intent-relevant annotations, making it hard for models to infer human intent from textual modifications. We introduce an intent-centric image–text dataset generated via reasoning by a Multimodal Large Language Model (MLLM) to better train ZS-CIR models for human manipulation intent understanding. Building on this dataset, we propose De-MINDS, a framework that distills the MLLM’s reasoning ability to capture manipulation intent and enhance models’ comprehension of modified text. A simple mapping network translates image information into language space and combines it with the manipulation text to form a query. De-MINDS then extracts intention-relevant information from this query and encodes it as pseudo-word tokens for accurate ZS-CIR. Across four ZS-CIR tasks, De-MINDS shows strong generalization and improves over existing methods by 2.15% to 4.05%, establishing new state-of-the-art results with comparable inference time.

High-precision path tracking of Autonomous Mobile Robots (AMRs) is challenging due to nonlinear dynamics, external stochastic disturbances, and drastic payload variations. Traditional fixed-gain controllers often fail to maintain performance across varying load conditions, leading to safety risks and efficiency losses. This paper proposes a novel dual-loop adaptive control architecture combining H-infinity control based on Adaptive Dynamic Programming (ADP) and Recursive Least Squares (RLS) estimation. The inner loop utilizes an ADP framework with actor-critic neural networks to data-drivenly approximate the solution of the Hamilton-Jacobi-Isaacs (HJI) equation, generating a robust optimal control law against worst-case external disturbances without requiring an exact system model. Simultaneously, the outer loop employs an RLS estimator to online identify time-varying parameters such as mass and friction coefficients, feeding these estimates back to the ADP controller to compensate for internal model mismatches in real-time. Simulation results conducted on a high-fidelity Gazebo/ROS2 platform under drastic payload variations (switching between 50 kg and 300 kg) demonstrate that the proposed method reduces tracking errors by 88% compared to classical PID and 82% compared to ADP-only baselines. Furthermore, it achieves exceptional load-invariant performance with a negligible error gap between empty and full loads, validating its effectiveness for high-reliability AMR operations in dynamic industrial settings.

Live streaming has gained widespread popularity for its rich interactive experience, generating massive volumes of multimedia network signals on the Internet. However, this medium has also been abused to disseminate harmful content, posing significant challenges to information forensics and network traffic analysis. Due to the real-time dynamics and segment homogeneity of live streaming, conventional video traffic identification methods are often ineffective. To this end, we propose LSTI, the first efficient and lightweight method for live streaming traffic identification. LSTI constructs live fingerprints by simulating the workflow of the HLS protocol and leverages time-domain and frequency-domain features for rapid identification of encrypted traffic. Experimental results show that LSTI achieves identification accuracies of 0.998 and 0.975 on high-volatility and low-volatility samples, outperforming existing baselines while exhibiting strong robustness to noise and cross-platform compatibility.

Network Covert Timing Channels (NCTCs) pose a serious threat to network security, through which attackers transmit hidden information by manipulating inter-packet delays (IPDs). Existing methods have shown strong performance in IPv4 networks by relying solely on timing features. However, more flexible routing mechanisms may introduce additional timing jitter into benign traffic in IPv6; meanwhile, the heterogeneous processing of optional IPv6 extension headers by routers make per-hop processing delays across routers increasingly unpredictable. These factors collectively increase the complexity of IPv6 IPD patterns, making timing-only detection methods insufficient for accurately modeling legitimate baseline behavior. To address such challenges, we propose TRANCTC, a Transformer-based anomaly detection method. By modeling temporal-structural alignment, TRANCTC effectively overcomes the limitations of timing-only baseline construction and learns a multidimensional representation of legitimate IPv6 traffic. On the public CAIDA dataset, TRANCTC significantly outperforms existing approaches across multiple NCTC types in the IPv6/TCP setting, and ablation studies show that incorporating structural features and contextual modeling substantially improve detection accuracy. TRANCTC fills the gap in detecting covert timing channels in IPv6 networks.

Grokking in modular arithmetic has established itself as the quintessential fruit fly experiment, serving as a critical domain for investigating the mechanistic origins of model generalization. Despite its significance, existing research remains narrowly focused on specific local circuits or optimization tuning, largely overlooking the global structural evolution that fundamentally drives this phenomenon. We propose that grokking originates from a spontaneous simplification of internal model structures governed by the principle of parsimony. We integrate causal, spectral, and algorithmic complexity measures alongside Singular Learning Theory to reveal that the transition from memorization to generalization corresponds to the physical collapse of redundant manifolds and deep information compression, offering a novel perspective for understanding the mechanisms of model overfitting and generalization.

Malicious domain detection is a key challenge in network security. Traditional methods cannot achieve effective malicious domain detection in real-world limited scenarios with limited labeled data and insufficient domain relationships, which we focus on in this paper. Based on the insight that domains with similar encrypted traffic behavior are expected to share similar representations in the embedding space, we propose MDDB-AETB, boosting malicious domain detection based on alignment with encrypted traffic behavior. We extract text and behavior features from TLS messages of encrypted traffic. For behavior features, we measure the similarity of statistical features as self-supervised learning labels. With these labels, we fine-tune the pre-trained model whose input is domain text, getting an optimized embedding representation model. The loss function for fine-tuning combines mean square error (MSE) loss and contrastive loss to capture the subtle behavior of encrypted traffic, enhancing its detection capability. We evaluate MDDB-AETB against three state-of-the-art baselines, the results show that MDDB-AETB consistently achieves the best performance across all test set proportions, reaching up to 99% F1-score while maintaining stable advantages even under limited training data.

Accurate encrypted traffic classification (ETC) plays a crucial role in network management and security. Existing methods rely on the assumption that complete bidirectional traffic is available for analysis. However, asymmetric routing and load balancing in real-world network environment may result in unidirectional traffic, thus leading to significant performance degradation. In this paper, we propose Dual-Graph, a protocol interaction-aware representation framework for accurate unidirectional ETC. Dual-Graph enhances flow representations by exploring the latent bidirectional interaction modes within the unidirectional traffic. Specifically, through our designed interaction feature extraction and pseudo-burst reconstruction, Dual-Graph learns to infer the missing bidirectional information from the given unidirectional flow, and further captures internal interaction patterns and dependencies via graph-based modeling, thereby generating effective representations for various downstream tasks. Extensive experiments demonstrate that Dual-Graph consistently outperforms state-of-the-art methods and exhibits strong robustness across diverse scenarios.

Digital light processing (DLP) technology offers great promise for fabricating geometrically intricate silicon nitride (Si3N4) ceramics, yet its advancement has been persistently constrained by the intrinsic challenge of formulating slurries that simultaneously enable high solid loading and low viscosity. This work presents a synergistic strategy that effectively resolves this bottleneck. Our approach integrates optimized physical packing (coarse/fine of 3:7) with precise colloidal interface engineering, employing a novel acrylate-modified block copolymer (KMT-3510) in conjunction with a short-chain dispersant (BYK-9076). Mechanistic insights reveal that the polar groups (C--O and ester C-O) of KMT-3510 form strong hydrogen bonds with the Si3N4 surface, while its hydrophobic alkane chains (C-H) extend into the resin matrix, establishing a robust steric hindrance barrier. This synergistic dispersion enabled a Si3N4 slurry with a high solid loading of 60 vol% and a low viscosity of 3.10 Pa & sdot;s at 12 s- 1, demonstrating excellent rheological performance. Using this high-solid-loading slurry, fully dense ceramic parts were successfully fabricated via DLP, followed by debinding and gas-pressure sintering at 1800 degrees C. The sintered ceramics inherited excellent mechanical properties, including a flexural strength (742 +/- 24 MPa), Vickers hardness (16.8 +/- 1.6 GPa), and fracture toughness (5.02 +/- 0.34 MPa & sdot;m1/2). Microstructural analysis indicated that the high toughness originates from the synergistic activation of multiple toughening mechanisms, such as crack deflection, grain bridging, and beta-Si3N4 grain pull-out. This work provides a clear and effective slurry design pathway for the DLP-based fabrication of high-performance Si3N4 ceramics with complex geometries.

Website fingerprinting attacks allow an eavesdropper to determine which websites a user is visiting by analyzing metadata from encrypted traffic, thereby posing a significant threat to online privacy. Most current methods rely on the assumption that a user visits only a single webpage at a time, which fails to account for the prevalence of multi-tab browsing in realistic scenarios. In such environments, traffic from multiple websites is interleaved, and practical attacks are further hindered by the scarcity of target-domain data, which limits the training efficacy and generalization of conventional models. To address these challenges, this paper proposes CDWF, a cross-domain few-shot website fingerprinting framework for multi-tab settings. The framework utilizes a three-stage progressive learning pipeline. First, traffic-augmented pre-training on public datasets extracts generalizable traffic representations through contrastive learning with diverse packet-level perturbations. Next, adversarial domain adaptation to align feature distributions between the source and target domains. Finally, few-shot target adaptation using a minimal set of target-domain samples and their augmented versions for rapid deployment. Experimental results demonstrate that CDWF significantly outperforms existing baselines in closed-world, open-world, and various defense scenarios. CDWF maintains superior robustness and recognition accuracy even when the number of tabs is unknown and under strong defensive conditions. This work provides an effective solution for practical website fingerprinting attacks in environments with limited data and constrained resources.

With the proliferation of anonymous networks, traffic correlation attacks have become a common means of deanonymization. However, existing methods often rely on regular traffic patterns and exhibit limited effectiveness under bursty traffic conditions typical of web interactions—a critical and unresolved challenge. To overcome this limitation, we propose EnhanCorr, an enhanced flow correlation method based on multimodal convolution, designed to improve de-anonymization performance in bursty traffic environments. Our approach introduces a hybrid input matrix and a window-based weighting mechanism to enhance correlation stability, alongside pre-matching and window-sampling strategies to accelerate training and boost inference efficiency. Experimental results demonstrate that EnhanCorr achieves superior correlation accuracy and computational efficiency in bursty traffic scenarios, confirming its practical value and applicability in real-world anonymous communication systems.

In recent years, deep learning-based website fingerprinting (WF) attacks have posed significant threats to anonymous communication networks such as Tor. Existing defenses struggle to balance low overhead with strong protection. In this paper, we propose Flash-Flood, a generative WF defense framework based on Consistency Models and temporal manifold reconstruction. It compresses multi-step diffusion into one-step inference and generates high-fidelity adversarial traffic to mask real sequences. In addition, a best-fit parasitic fusion mechanism embeds real communication into pre-generated cover timelines, enabling global topology-level feature reshaping. Experimental results show that, with only 13.14 ms generation latency and approximately 34.81

In indirect-drive inertial confinement fusion research, precise diagnostics of ablator electron temperature evolution are essential for understanding radiative ablation behavior. Using silicon-traced CH samples with point-projection backlighting, we measured time-resolved backlight spectra and silicon plasma absorption spectra at different times, deriving transmission spectra. Radiation temperature on sample was determined via the 3D view-factor code IRAD3D, while radiation-hydrodynamic simulations provided the evolution of electron temperature and density in the silicon plasma. By comparing experimentally measured transmission spectra with theoretically calculated spectra at varying electron temperatures, we inferred the electron temperature of the silicon plasma. Results reveal a rise-then-fall electron temperature trend in the silicon layer, with agreement between experiment and simulation during the temperature decline phase but discrepancies in the rise phase due to ionization-state complexities. This work elucidates electron temperature evolution and transport mechanisms during radiative heat wave propagation in low-Z materials.

Encrypted traffic classification is essential for network security, management, and Quality of Service (QoS), but models trained under IID assumptions often degrade after deployment due to application updates, temporal drift, and infrastructural changes. We observe that short-window multi-flow context provides more stable behavioral cues than isolated flows, while self-supervised pre-training can learn transferable priors from unlabeled traffic without target-domain labels. We present Odysseus, a context-level pre-training framework for OOD encrypted traffic classification. Odysseus converts raw pcaps into a Multi-Scale Flow Grid Representation (MSFGR) and pre-trains a masked autoencoder with complementary packet- and flow-level masking to capture intra-/inter-flow structure. During fine-tuning, it integrates Hierarchical Attention Distillation (HAD), Traffic-Aware Augmentation and Contrastive Learning (TACL), and Associated Flow Prediction (AFP) to enhance context modeling, distill context knowledge into a deployable flow-level branch, and regularize OOD-robust relational representations. Across 11 scenarios (5 IID, 6 OOD) spanning Cross-Version, Cross-Time, Cross-Spatiotemporal, and Cross-Type shifts, Odysseus achieves state-of-the-art performance and improves OOD F1-Score by up to $15.70\%$, reaching $85.38\%$, $90.05\%$, $71.93\%$, and $99.62/96.87/72.55\%$ F1 on the corresponding OOD settings. Further analyses confirm the effectiveness of each component, strong few-shot robustness, graceful degradation under noisy/missing context, and advantages over matrix-based inputs.