We consider a non-Bayesian estimation problem preceded by a detection step. In this framework, the observation model follows one of two hypotheses: an informative hypothesis, where the data contain information about the parameters of interest, and a non-informative hypothesis, where no such information is present. Estimation is only performed if the detection stage declares the informative hypothesis. This setup is common in various signal processing applications, such as communication systems, where decoding occurs only upon message detection, and radar systems, where direction-of-arrival estimation follows target identification. Since the estimation errors are well-defined only when the estimation process occurs, the standard mean-squared error (MSE) criterion is inappropriate for this post-detection estimation problem. To address this, we leverage estimation theory under misspecified models. First, we present the post-selection pseudo-true parameter. Then, we introduce the misspecified post-selection MSE (MPSMSE) criterion, which quantifies the MSE between the estimator and the post-selection pseudo-true parameter conditioning on the event of detecting the informative hypothesis by the considered detector. We developed a new Cramér-Rao-type bound on the MPSMSE that incorporates the knowledge on the detection stage with potential model misspecification. This bound is based on the well-known misspecified Cramér-Rao bound (MCRB), and it provides a robust and accurate benchmark on the MPSMSE for analyzing the performance of practical estimators. Simulations confirm that the proposed bound is a valid lower bound on the MPSMSE of practical estimators and is tighter than existing alternatives.
This paper introduces a framework for spectral analysis of weak signals in the presence of strong interference that can potentially exceed the dynamic range of the analog-to-digital converter (ADC) employed for data acquisition. The framework combines modulo sampling to avoid ADC saturation with interference cancellation (IC) and adaptive spectral estimators such as amplitude and phase estimation of a sinusoid (APES) and Capon for high-resolution spectral analysis. Numerical simulations validate the proposed approach’s ability to accurately recover weak signals’ spectrum with significantly lower mean-squared error (MSE) in amplitude estimation compared to conventional methods in the considered scenarios. Notably, it is shown that APES and Capon are able to effectively reject the out-of-range interference and reveal weak spectral content, without requiring the interference cancellation step.
This paper discusses a method for classification of breast cancer imaging data through the application of an adaptive neuro-fuzzy inference system (ANFIS) and particle swarm optimization (PSO) for hyperparameter optimization of the ANFIS system. A robust parameter tuning method is used to select the optimal configuration for the ANFIS and PSO components without expert knowledge of the dataset. Using these methods, high classification accuracies can be achieved for both the original and diagnostic versions of the Wisconsin Breast Cancer Dataset. These results demonstrate the flexibility and potential of a joint ANFIS-PSO system for automated diagnosis while retaining system simplicity and linguistic interpretability to support clinical decision-making.
Use of millimeter waves (mmWave) and other high frequency bands are expected to be a crucial part of 6G networks. Performance analysis of mmWave communication scenarios and modeling of various protocols in those situations have demonstrated the burst-error nature of mmWave channels. With the increasing adoption of such communication standards operating in lossy conditions, advanced communication techniques have started to be explored within the context of bursty channels. Network coding is one such widely utilized technology for achieving high-throughput and reliable communication over lossy channels. This work focuses on modeling the TCP and TCP/NC protocols within a burst-error scenario, mainly focusing on the presence of time-out events as sources of window size reduction. We provide a detailed analysis of throughput and congestion control, along with their formulation and simulated results. We show that, even with the presence of time-out events, TCP/NC outperforms standard TCP by over an order of magnitude.
This article presents a comprehensive review of recent advances in intrinsic Cramer-Rao bounds (ICRBs) for Lie groups (LGs), which play a pivotal role in addressing estimation problems involving parameters and/or observations constrained by geometric structures. The review encompasses both deterministic and Bayesian frameworks, with a detailed examination of their formulation, derivation, and theoretical foundations. Furthermore, we underscore significant theoretical contributions and extend the discussion to practical estimation challenges, offering insights into their applicability. Emphasis is placed on methodologies for validating these bounds, providing a robust framework for performance evaluation across a variety of estimation problems in engineering and applied sciences.
Capsule networks offer a useful approach for modeling part-whole hierarchies in visual data, yet they remain limited in effectively handling uncertainty in part-object relationships. This work introduces an entropy-adjusted dynamic routing (EADR) algorithm that leverages information-theoretic principles to enhance both the performance and interpretability of capsule networks. By incorporating an entropy-based regularization term into the final iteration of the dynamic routing process, our approach refines routing decisions, reducing reliance on uncertain capsule connections. Experimental evaluations on the CIFAR-10 dataset demonstrate that our method achieves a mean accuracy of 85.73%, surpassing the baseline accuracy of 84.22% achieved by standard dynamic routing. Comparative analysis with recent entropy-based routing methods highlights our approach’s balance of computational efficiency and routing flexibility, achieved without much additional model complexity. These results suggest that EADR routing can be a useful tool for enhancing both the interpretability and efficacy of capsule networks in uncertain environments.
As electric vehicles (EVs) gain popularity, the demand for efficient EV routing and innovative charging solutions is increasing. This study proposes a novel integration of Temporal Multimodal Multivariate Learning (TMML) and the Time-Dependent Shortest Path (TDSP) algorithm to dynamically optimize EV routing while incorporating stationary and wireless charging systems. TMML continuously updates travel time distributions based on real-time observations, reducing uncertainties and enabling more accurate predictions. These updated distributions are utilized by TDSP to determine the most efficient routes, considering time-dependent travel times, and the availability of charging infrastructure. The methodology integrates Mobile Energy Distributors (MEDs) and Dynamic Inductive Charging (DIC). MED dynamically dispatches portable charging units, while DIC enables on-the-move charging through road-embedded systems. Real-world data from the Washington, DC metropolitan area, sourced from Regional Integrated Transportation Information System (RITIS.org), provides granular insights into traffic patterns and charging infrastructure performance. This hybrid system operates by extracting feasible routes, evaluating them using TMML and TDSP, and selecting the optimal path to minimize travel time, waiting time, and charging costs.
Decentralized federated learning (DFL) offers enhanced resilience to client failures and potential attacks than its centralized counterpart. This advantage stems from its ability to aggregate learning models from distributed clients requiring centralized server coordination. However, the practical adoption of DFL faces several challenges that threaten the robustness of local models. On one hand, the distribution of data might change over time, degrading the aggregated model’s performance on test data. On the other hand, Byzantine attacks, where certain users send malicious updates to their neighbors to spread erroneous knowledge, can compromise the convergence and accuracy of the global model. Notably, no existing work has simultaneously addressed both distributional shifts and Byzantine attacks in decentralized settings. To bridge this gap, we first propose a robust aggregation algorithm, Local Performance Evaluation with Temperature-Scaled Softmax Reweighting (LPE-TSR), to defend against Byzantine attacks. We then integrate Wasserstein distributionally robust optimization with LPE-TSR and develop Distributional and Byzantine Robust Decentralized Stochastic Gradient Descent (DB-Robust DSGD) to tackle both challenges simultaneously. DB-Robust DSGD allows flexible selection of robust aggregation algorithms tailored to specific scenarios. Experimental results show that LPE-TSR achieves optimal performance across diverse attack scenarios, while DB-Robust DSGD effectively mitigates both distributional shifts and Byzantine attacks.
Digital twins (DTs) are virtual representations of physical systems, replicating their behavior, dynamics, and interactions with the environment. Hosted on computational platforms, DTs have access to more extensive information and computational power than their physical counterparts (PTs), making them suitable for control applications. This paper proposes a novel DT-based control architecture that leverages causal learning to achieve autonomous control. In this framework, the DT employs causal inference to determine the desired system behavior and, subsequently, instructs the PT accordingly. By incorporating causal learning, a DT can gain self-training capabilities, enabling it to generate and analyze hypothetical scenarios based on cause-and-effect relationships. Simulation results show that DT-based control outperforms the traditional local control method by 49% in terms of the tracking performance measured through the mean squared error. Additionally, causal learning demonstrates 16% better tracking performance over statistical learning in a self-training scenario, as measured by the mean squared error.
The extraction of electrical network frequency (ENF) data from audio signals has become a key tool in multimedia forensics, enabling applications such as timestamping, authentication, and geolocation estimation. In particular, timestamping of audio recordings can be performed by extracting the ENF signal and correlating the result with reference records. In this paper, we present a comparative analysis of ENF-based methods for accurate timestamping of audio recordings using real-world data from various sources. We analyze the accuracy of these methods in estimating the timestamps of the records. We compare the influence of different parameters, such as the duration of the target signal and the reference signal, and the use of different correlation metrics. In addition, we provide a robust platform for the empirical evaluation of the ENF extraction methods and the features of the target-reference correlation approach. Our results offer several insights and practical recommendations for optimizing ENF-based timestamping approaches.
In statistical signal processing and estimation theory, discrepancies between the true data-generating model and the assumed model can lead to large estimation errors. The misspecified Cramér-Rao bound (MCRB) quantifies the impact of model mismatch on the mean-squared error (MSE), but it is derived for the asymptotic region, where the estimation errors are small. Consequently, it cannot be used to investigate estimation performance in the non-asymptotic region, where estimation errors are large. For instance, the MCRB is not effective in predicting the threshold phenomenon, which is crucial in many signal processing applications. The Barankin bound offers a tighter bound in the non-asymptotic region, and several works have demonstrated its applicability in predicting the threshold phenomenon. However, it is derived for perfectly specified models, where the true data-generating model matches the assumed model. In this work, a misspecified Barankin-type bound that accounts for model mismatch, enabling investigation of the impact of misspecification on the threshold phenomenon, is derived. The Barankin bound and the MCRB emerge as special cases of this lower bound. To illustrate its utility and investigate threshold signal-to-noise ratio phenomenon, we apply the proposed bound to the problem of direction-of-arrival estimation using a sensor array under model misspecification. The modeling errors induce large errors and distort the ambiguity function, a behavior effectively captured by the proposed bound.
Independent Component Analysis (ICA) is a powerful data-driven method that has been widely applied in functional magnetic resonance imaging (fMRI) data analysis to uncover underlying sources. An attractive way to boost ICA performance is via constraints to guide ICA factors to be similar to user-supplied "references", allowing incorporation of prior-knowledge into the factorization. However, most of existing constrained ICA methods typically only impose source constraints and are unable to impose constraints on the mixing matrix. With multi-subject medical imaging datasets, constraining the mixing matrix with subjects’ symptom-related measurements, such as clinical scores or cognitive variables, enhances the algorithm’s ability to identify brain activities associated with these symptoms. This offers a novel perspective for understanding the pathologies underlying various psychiatric disorders. Therefore, to overcome the limitations of existing constrained ICA algorithms, we introduce a new constrained ICA algorithm: adaptive-reverse constrained matrix entropy bound minimization (arc-M-EBM), which imposes constraints on the mixing matrix and uses adaptive-reverse thresholding to avoid overfitting or underfitting. This approach ensures flexibility and leads to more accurate and interpretable source separation. Simulations demonstrate that arc-M-EBM outperforms traditional ICA methods. Application to resting-state fMRI data from 176 subjects from healthy controls and patients reveals significant relationships between constrained components and clinical measures, enhancing our understanding of brain-behavior relationships.
Today’s digital world facilitates the rapid growth of social media platforms and online news sources and consequently the spread of true and false information. While the detection of fake news is extensively studied using classical machine learning (ML)-based approaches, quantum ML approaches remain largely unexplored. In this study, we aim to investigate the effectiveness of a hybrid quantum-classical neural network on three open access datasets. This study proposes a Hybrid Quantum Neural Network (HQNN) model for classifying news articles as either "fake" or "real". The approach combines classical machine learning techniques with quantum computing to leverage the strengths of both paradigms. It incorporates a quantum layer, implemented via a parameterized quantum circuit that utilizes angle embedding and entanglement to extract complex features, followed by a classical fully connected layer for final decision-making. Our results demonstrate the effectiveness of the HQNN model, achieving high accuracy across multiple datasets, with values reaching up to 90.71%. These findings highlight the potential of hybrid quantum-classical models to improve fake news detection and pave the way for further research into quantum-enhanced ML applications.
As remote work becomes more common, organizations face challenges in ensuring compliance with information security policies (ISP) among their decentralized employees. This paper explores the role of autonomy, using Self-Determination Theory (SDT), to understand its impact on compliance intentions regarding ISPs in flexible work environments (FWA). Data collected from 114 employees in flexible work settings revealed a strong positive relationship (rs = .96, p < .001) between autonomy and compliance intentions. The findings highlight the importance of fostering autonomy to enhance intrinsic motivation, improve adherence to ISPs, and create a resilient organizational security culture. The paper also discusses practical implications for organizations leveraging autonomy in FWA.
The rapid development of large language models (LLMs) has significantly increased their engagement frequency in our daily lives. Due to the large amount usage of LLM agents, users are beginning to consider whether agents can serve as reliable and cooperative assistants for their benefit. To this end, game theory, as an effective tool in the study of strategic interactions, has gathered attention and has been employed in the research field of LLMs, particularly in exploring their interactions with users. Most previous studies focused on the performance of LLMs in static games or finitely repeated games, but these studies are relatively simplistic and fail to fully capture the characteristics of User-LLM interactions. To address the issues, our research proposes a novel approach using infinitely repeated games to analyze LLMs’ behavioral traits. Bayesian inference is incorporated by providing probabilistic information to assist LLMs in their decision-making process, using the beta distribution as both the prior and posterior for consistency. We conduct a case study over the trending LLMs: GPT3, GPT4 and Llama3. Experimental results demonstrate that LLMs outperform the mixed Nash equilibrium in infinitely repeated games, indicating their capability for cooperation in repeated interactions. The integration of Bayesian inference shows that LLMs can effectively process probabilistic information and improve their performance by an average of 5.26%. Our findings suggest that LLM agents have the preference to consider future payoffs rather than only caring about single-stage rewards, as well as the ability to build and maintain long-term cooperative relationships with users.
This paper introduces a novel approach to minimizing the Mean Squared Error (MSE) in channel estimation between mobile users and Reconfigurable Intelligent Surfaces (RIS) within wireless communication systems. Our model includes a mobile user, an RIS with multiple elements, and a base station (BS) with multiple antennas. We apply the Kalman Filter (KF) and Extended Kalman Filter (EKF) like algorithms to reduce MSE in a non-linear, distance-dependent channel model. Additionally, we propose a Non-Circular Noise Kalman Filter (NCNKF) to address scenarios with non-circular complex state-space noise. Results show that KF and EKF can achieve lower MSE in estimating the channel than other known approaches, while the NCNKF algorithm outperforms others in non-circular state-space noise environments. The study concludes with numerical comparisons and an in-depth discussion of the performance improvements enabled by our approach.
Yoga is widely recognized for its physical and mental health benefits, but existing resources like video tutorials, mobile apps, and online classes often lack personalized feedback, making it difficult to ensure proper posture and avoid injury. For individuals with disabilities or limited mobility, these challenges are even greater, as traditional resources may not cater to their specific needs. Previous studies show that incorrect posture during yoga can increase the risk of discomfort or injury, highlighting the importance of proper alignment and personalized feedback for safe practice. In this paper, we introduce the virtual yoga instructor application which consists of a virtual instructor and a feedback module via pose estimation. The virtual instructor, implemented in Unity game engine with an animated 3D humanoid model, demonstrates yoga poses. The feedback module uses pose estimation to detect and analyze human body landmarks, generating scores and providing feedback through text and visual markers to guide users in achieving accurate yoga postures. The conducted user evaluations assess the application performance based on the criteria of ease of use, interaction, informativeness, engagement, and retention. The results indicate that our proposed application was highly rated and consistently preferred by users.
This paper proposes a methodology that enables physical resource blocks (PRBs) in 5G networks to be overbooked among multiple users while diligently satisfying their Quality of Service (QoS) requirements. Our overbooking methodology is guided by a novel optimization framework, which is solved using a Genetic Algorithm (GA). It is shown that, under inherent uncertainties associated with communication processes, our overbooking methodology enhances important performance criteria, such as resource usage efficiency and the number of supported users, beyond what is permitted by traditional approaches that do not allow overbooking. Numerous simulation results under varied considerations have been presented, which show the performance advantages of our proposed mechanism.
In this paper, a deceptive signal-assisted private split learning is investigated. In our model, several edge devices jointly perform collaborative training, and some eavesdroppers aim to collect the model and data information from devices. To prevent the eavesdropper from collecting model and data information, a subset of devices can transmit deceptive signals. Therefore, it is necessary to determine the subset of devices used for deceptive signal transmission, the subset of model training devices, and the models assigned to each model training device. This problem is formulated as an optimization problem whose goal is to minimize the eavesdropped information while meeting the model training energy consumption and delay constraints. To solve this problem, we propose an actor-critic deep reinforcement learning (AC-DRL) framework that enables a centralized agent to dynamically determine the model training devices, the deceptive signal transmission devices, the transmit power, and sub-models assigned to each model training device while considering network topology and device properties (e.g., energy and privacy constraints). The proposed AC-DRL framework integrates an actor network to generate optimal actions and a critic network to stabilize the learning process by evaluating the cumulative rewards. Simulation results demonstrate that the proposed method improves the convergence rate by 3x and the accumulated reward by up to 40% compared with the proximal policy optimization (PPO) algorithm.
The generalization of the AM-GM inequality for matrices helps in understanding the difference between with-replacement sampling and without-replacement sampling in stochastic gradient descent (SGD). However, it was previously shown that the matrix form of the AM-GM inequality does not hold for sets of five or more matrices [1], [2]. An open problem presented in COLT 2021 [3] suggested that this limitation may not apply to matrices with small condition numbers, proposing this as a conjecture. In this paper, we establish a weak matrix AM-GM inequality for positive-semidefinite matrices with small condition numbers. Our findings indicate that the performance of SGD without replacement can not be significantly worse than with-replacement sampling, while the converse does not hold.
