Deploying a data-driven model in a real environment is challenging due to the fact that the distribution of the input data is usually not the same as that of the training data. The distribution change can considerably degrade the reliability and robustness of the model output. In this study, we develop a chance constrained back-mapping approach to transform the features of input data with a varying distribution into the distribution of the training dataset, based on the Kantorovich transportation. Since the Kantorovich method cannot guarantee a truthful backmapping of samples with highly variable size and distribution, we introduce a constraint to restrict the transformation by ensuring a prospect certainty of high accuracy of the model output. In addition, to reflect the data variability, the upper bound of the constraint is considered as a random parameter whose distribution is estimated online using the rolling average method. Furthermore, chance-constrained optimization is used to satisfy this constraint in a probabilistic manner. Therefore, the proposed approach enssures a high certainty of high accuracy of the trained model when deployed in a variable environment. Both analytical and experimental verification of the approach are carried out. The results of synthetic and real datasets demonstrate a clear improvement in terms of accuracy and robustness compared with the existing methods.

Diffusion models have become a cornerstone of generative modeling, excelling in tasks such as image creation, text synthesis, audio generation, and scientific simulations. These models rely on iterative refinement processes to generate high-quality outputs, but their computational demands remain a significant bottleneck. The need for hundreds or thousands of sampling steps often limits their practicality in real-time applications, resource-constrained environments, and energy-efficient systems. In recent years, extensive efforts have been made to enhance the efficiency of diffusion models, addressing their inherent computational complexity. Key advancements include the development of faster sampling algorithms, model compression techniques like pruning and quantization, and knowledge distillation strategies to reduce model size without compromising quality. Additionally, the integration of hardware-specific optimizations, such as GPU and TPU acceleration, has enabled diffusion models to perform efficiently on both cloud platforms and edge devices. This survey systematically examines the state-of-the-art techniques for improving the efficiency of diffusion models. It categorizes these innovations into three primary areas: sampling acceleration, model optimization, and hardware-aware design. Furthermore, it highlights the transformative impact of efficient diffusion models in applications ranging from creative industries and real-time systems to scientific research and sustainable AI. By analyzing current trends and identifying open research directions, this survey aims to provide a roadmap for advancing the scalability and accessibility of diffusion models across diverse domains.

This paper presents a conformal aperture inverse design framework based on the inverse design method based on spectral analysis (IDMBSA) to address efficiency and applicability challenges in conformal aperture design. The framework initially employs IDMBSA to devise planar aperture field distributions, subsequently extending them to curved surfaces through full-wave simulations based on the equivalence principle and uniqueness theorem. Furthermore, we examine the influence of field truncation on the implementation of the equivalence principle. The results demonstrate that this framework effectively reduces the dimensionality of conformal array design problems, significantly improving efficiency for conformal aperture design. The feasibility of the method is validated through simulations, prototype fabrication, and testing for sparse conformal arrays. The innovation of this paper lies in combining IDMBSA with the equivalence principle, proposing a versatile and efficient conformal aperture inverse design framework. This approach not only handles various complex curved surfaces but also substantially enhances design efficiency, offering new solutions for conformal antenna applications in aerospace and other fields.

Risk analysis is a key process in project management, aiming to evaluate and mitigate risks by quantifying the probability and severity of potential events. A common tool for this purpose is the risk matrix, which offers a structured visualization of risks. Despite its effectiveness, constructing risk matrices traditionally depends on domain experts to define event probabilities and severities, leading to scalability and consistency challenges. This paper introduces RMGNN (Risk Matrix Generation with Graph Neural Networks), a semi-supervised and transductive method for automating risk matrix construction. RMGNN models historical events as graph structures, leveraging graph neural networks and language models to estimate event probabilities and severities. The method integrates labeled and unlabeled data, reducing reliance on manual annotation while maintaining accuracy. The main contributions of this paper include: (1) a graph-based framework for semi-supervised and transductive risk matrix learning, (2) a unified approach for estimating event probability and severity from graph representations, and (3) the use of graph topology to enhance interpretability in risk assessment. Experiments on real-world datasets demonstrate the effectiveness of RMGNN in addressing challenges associated with traditional risk matrix construction, supporting improved risk analysis and decision-making processes.

Traditional banking systems rely heavily on centralized data storage and identity verification processes, leading to security vulnerabilities and regulatory burdens. Zero-Knowledge Proof (ZKP) technologies offer a cryptographic solution that enables financial verification-such as identity authentication, balance confirmation, and transaction validation-without disclosing sensitive user data. This paper explores the integration of zk-SNARK and zk-STARK protocols, ZK-Rollup-based scalability solutions, and zkLedger applications within banking operations. It examines the role of selective disclosure mechanisms in critical areas such as KYC/AML compliance, balance verification, cross-border transactions, and regulatory audits. Furthermore, a four-layer architectural model is proposed to facilitate the seamless integration of ZKP-based frameworks into traditional banking infrastructures, addressing key aspects such as security optimizations, transaction costs, and scalability. Lastly, the potential of ZKP in regulatory compliance, its bridging role between DeFi and centralized banking, and the emergence of ZKP-driven next-generation banking services are discussed.

We propose Supervised Learning Preference Optimization (SLPO), an approach to aligning language models that relies solely on the classic supervised learning loss function, cross-entropy. Unlike Reinforcement Learning from Human Feedback (RLHF), SLPO directly adjusts probability mass for chosen, rejected, and other sequences relative to a reference model, eliminating the need for KL divergence regularization. SLPO also avoids the extensive reparameterization of model outputs into scores and the Bradley-Terry model required by Direct Preference Optimization (DPO). As a result, SLPO is a purely supervised learning approach to alignment, free of reinforcement learning concepts. We further demonstrate how to efficiently implement a targeted probability mass for the intractably large set of sequences that are neither explicitly chosen nor rejected.

5G and beyond provides connectivity for a variety of heterogeneous, often mission critical services, placing stringent performance requirements on these systems. Providing satisfactory Quality-of-Experience (QoE) for diverse, coexisting applications prompts the network operators to enforce application-aware, efficient resource allocation schemes that can increase user-satisfaction, system utilization, and decrease the operational costs. For these purposes, QoS Flows and network slicing have been identified as key enablers. Those concepts move away from economy of scale, towards a fine-grained slice and flow handling with customized resource control for each application, application type, or slice. Implications of this change on the scaling of network resources and how to optimally allocate the available resources have been largely ignored so far. Furthermore, while capacity has been recognized as a key resource, selecting the appropriate queue size, granularity of the resource allocation scheme, and their relations with the number of clients are often neglected in the process of resource dimensioning. To address these shortcomings, we perform an in-depth evaluation of the effects that impact factors have on the overall QoE and system utilization using the OMNeT++ simulator. We show the optimization potential for QoE and resource utilization, and further formulate guidelines for efficient and QoE-aware resource allocation.  

Probabilistic Neural Contextualization introduces a groundbreaking approach that bridges probabilistic reasoning with neural architectures to redefine contextual knowledge retrieval in computational models. By integrating dynamic contextualization mechanisms, it achieves enhanced adaptability and precision in managing uncertainty across diverse linguistic domains. The novel architecture incorporates advanced probabilistic inference layers, ensuring that outputs are consistently aligned with the complex demands of varied contextual scenarios. Comprehensive evaluations demonstrated marked improvements in key performance metrics, including perplexity and F1-score, validating the model's robustness and scalability. Experimental results further revealed its computational efficiency, maintaining performance consistency even under significant data variability and noise. Through seamless integration with open-source frameworks, the proposed methodology demonstrates its practical viability for large-scale applications, addressing critical challenges in knowledge-driven processing tasks. The study highlights a transformative advancement in contextual modeling, paving the way for more reliable, efficient, and intelligent computational systems capable of operating in complex and dynamic environments.

Potential for subsynchronous interactions (SSI) involving inverter-based resources is often screened for using small-signal impedance scans. However, these impedance scans are often slow to compute and it is often desired to study a wide range of operating conditions. This paper presents modeling practices and a new scanning approach for performing impedance-based SSI screening studies that can reduce the required computation time by orders of magnitude with negligible effect on accuracy. Relevant principles are illustrated using a type3 wind turbine model, using two different reactive power control strategies that have significantly different SSI behavior. The discussion of modeling practices includes per-unitization, mechanical dynamics, timestep selection, PLL design, and negative-sequence behavior and covers both time-saving simplifications and pitfalls or unsuitable simplifications to avoid. The new scanning approach involves identifying sets of input frequencies that do not interfere with one another’s outputs and using crest factor minimization to develop multisine test signals.

This paper focuses on Ambimorphic prototype research in the field of information engineering. Firstly, it elaborates the research background, purpose and significance, and analyzes the concept of Ambimorphic prototype and related theoretical foundations. It then introduces in detail the application of public key-private key simulation in information security and communication systems, as well as the design and implementation of telephone semantic tools, including requirements analysis, architecture design and key technologies. Through experiments, performance tests and result analyses are conducted on both, including individual tests and comprehensive application tests. Finally, it summarizes the research results, limitations, and looks forward to future research directions, providing new ideas and methods for the development of information engineering.

Functional near-infrared spectroscopy (fNIRS) is a preferred neuroimaging technique for studies requiring high ecological validity, allowing participants greater freedom of movement. Despite its relative robustness against motion artifacts (MAs) compared to traditional neuroimaging methods, fNIRS still faces challenges in managing and correcting these artifacts. Many existing MA correction algorithms notably lack validation on real data with ground-truth movement information. In this work, we combine computer vision, ground-truth movement data, and fNIRS signals to preliminarily characterize the association between specific head movements and MAs. Fifteen participants (age = 22.27 ± 2.62 years) took part in a whole-head fNIRS study, performing controlled head movements along three main rotational axes. Movements were categorized by axis (vertical, frontal, sagittal), speed (fast, slow), and type (half, full, repeated rotation). Experimental sessions were video recorded and analyzed frame-by-frame using the SynergyNet deep neural network to compute head orientation angles. Maximal movement amplitude and speed were extracted from head orientation data, while spikes and baseline shifts were identified in the fNIRS signals. Results showed that head orientation and movement metrics extracted via computer vision closely aligned with participant instructions. Additionally, repeated and Up/Down movements tended to compromise fNIRS signal quality. The occipital and pre-occipital regions were particularly susceptible to MAs following Up/Down movements, while temporal regions were most affected by bendLeft/bendRight and Left/Right movements. These findings underscore the importance of cap adherence and fit in the relationship between movements and MAs. Overall, the work lays the foundation for an automated approach to developing and validating fNIRS MA correction algorithms.

Movable antenna (MA) has attracted increasing attention in wireless communications recently. As compared to conventional fixed-position antennas (FPAs), the geometry of MAs can be dynamically reconfigured, such that more flexible beamforming can be achieved for different purposes. In this paper, we investigate the application of MAs to wide-beam coverage, aiming to jointly optimize the MAs' beamforming weights and positions within a line segment to maximize the minimum beam gain among all possible directions in a target region. However, the resulting optimization problem is non-convex and difficult to be optimally solved. To tackle this difficulty, we first derive a closed-form optimal solution to this problem in the special case with two MAs. While for the case with more than two MAs, an alternating optimization (AO) algorithm is proposed to obtain a high-quality suboptimal solution, where the MAs' beamforming weights and positions are alternately optimized by applying the successive convex approximation (SCA) technique. To reduce computational complexity, we further propose a more efficient MA position optimization method by leveraging the frequency modulation continuous wave (FMCW) design. Specifically, we construct a spatial FMCW-based continuous phase profile for the entire line segment and then select an optimal set of MA positions to optimize the wide-beam coverage performance with their FMCW-based phase profiles, thus greatly simplifying the wide-beam design. Furthermore, we extend the proposed AO and FMCW-based algorithms for the linear MA array to the planar MA array. Numerical results show that both our proposed algorithms can significantly outperform conventional FPAs even with optimized beamforming weights.

This paper analyzes the performance of Alamouti coded short-packet non-orthogonal multiple access (NOMA) systems with hybrid multicast and unicast transmission over Nakagami-m fading, where only the statistical channel state information is available at the transmitter, and the multicast and unicast signals are intended for all users and a particular user, respectively. Due to some practical limitations, channel estimation errors (CEE) caused by a linear minimum mean-square error estimator and residual hardware impairments (RHI) caused by imperfect hardware implementation are taken into account. We first derive the approximate closed-form expressions for the average block error rate (BLER) and the corresponding asymptotic expressions at all users. Using such expressions, we analyze the diversity performance including conventional diversity order and finite-SNR diversity order. After this, we quantify the relationship among the common blocklenth, power allocation, and pilot sequence length under users' reliability constraints. Finally, numerical and simulation results verify our theoretical analysis and show that i) CEE and RHI greatly affects the transmission blocklength, ii) RHI leads to the error floor phenomenon at high SNRs and the finite-SNR diversity order is the effective performance metric in the presence of RHI, iii) there exist the optimal power allocation coefficients and pilot sequence length to minimize the transmission blocklength in the considered hybrid multicast-unicast system, and iv) the NOMA scheme is superior to the OMA counterpart by achieving low-latency transmission.

The deployment of IoT platforms for SmartCity applications is demanding solutions to assure security and integrity levels. In this context, physical unclonable functions (PUF) may overcome the security drawbacks of storing the security key in a non-volatile memory. The existence of SRAM in embedded systems has driven the implementation of PUF solutions using memory circuit as entropy sources. The use of a non-specific memory design requires the need of identify the adequate cells useful for PUF applications. This work proposes a new methodology to distinguish the adequate cells based on mismatch factor calculation. The mismatch analysis using physical and performance parameters shows the viability of the selection methodology and computes the robustness of the responses in terms of probability of error in the start-up value.

The rapid evolution of software development practices, coupled with the increasing prominence of open-source software (OSS) projects, has positioned versioning systems as invaluable repositories of information for analysis. This paper delves into the factors that shape software development activities by examining documented methodologies and tools used to mine and analyze OSS repositories. Through a detailed review of approaches addressing developer contributions, code modifications, and collaborative workflows, we identify key influences on productivity, code quality, and team dynamics. Our exploration includes an analysis of analytics techniques and tools, such as CVSSearch and CVSAnalY, that facilitate a deeper understanding of developer behaviour and project evolution. Furthermore, we discuss the implications of these insights for enhancing software engineering practices across both OSS and proprietary environments. This paper provides a conceptual framework that connects theoretical insights with practical approaches, offering actionable strategies to improve processes, foster collaboration, and support learning in software development. Ultimately, the findings hold potential value not only for practitioners aiming to optimize workflows but also for educators leveraging OSS environments to teach and evaluate software engineering concepts.

Feature discrimination is a crucial aspect in neural network design, as it directly impacts the network's ability to distinguish between classes and generalize across diverse datasets. The accomplishment of achieving high-quality feature representations ensures high intra-class separability and poses one of the most challenging research directions. While convetional deep neural networks (DNNs) rely on complex transformations and very deep networks to come up with meaningful feature representations, they usually require days of training and consume significant energy amounts. To this end, spiking neural networks (SNNs) offer a promising alternative. SNN's ability to capture temporal and spatial dependencies renders them particularly suitable for complex tasks, where multi-modal data are required. In this paper, we propose a feature discrimination approach for multi-modal learning with SNNs, focusing on audiovisual data. We employ deep spiking residual learning for visual modality processing and a simpler yet efficient spiking network for auditory modality processing. Lastly, we deploy a spiking multilayer perceptron for modality fusion. We present our findings and evaluate our approach against similar works in the field in classification challenges. To the best of our knowledge, this is the first work investigating feature discrimination in SNNs.

In this work, we investigate the underlying physical mechanism for electric-field induced resistive switching in Au/MoS 2 /Au based memristive devices by combining reactive Molecular Dynamics (MD) and ab-initio quantum transport calculations. Using MD with Au/Mo/S ReaxFF potential, we observe the formation of realistic conductive filament consisting of gold atoms through monolayer MoS 2 layer when sufficient electric field is applied. We furthermore instigate the rupture of the gold atom filament when a sufficiently large electric field is applied in the opposite direction. To calculate the conductance of the obtained structures and identify the High Resistance (HR) and Low Resistance (LR) states, we employ the ab-initio electron transport calculations by importing the atomic structures from MD calculations. For single-defect MoS 2 memristors, the obtained LRS, HRS current densities are in order of 10 7 A/cm 2 which agrees reasonably well with the reported experiments.

In Deep Neural Networks (DNN), the depth-wise separable convolution has often replaced the standard 2D convolution having much fewer parameters and operations. Another common technique to squeeze DNNs is heterogeneous quantization, which uses a different bitwidth for each layer. In this context we propose for the first time a novel Reconfigurable Depth-wise convolution Module (RDM), which uses multipliers that can be reconfigured to support 1, 2 or 4 operations at the same time at increasingly lower precision of the operands. We leveraged High Level Synthesis to produce five RDM variants with different channels parallelism to cover a wide range of DNNs. The comparisons with a nonconfigurable Standard Depth-wise convolution module (SDM) on a CMOS FDSOI 28-nm technology show a significant latency reduction for a given silicon area for the low-precision configurations.