This paper studies hallucinations in structured extraction using prompt-based Semantic Role Labeling (SRL) as a controlled case study. We implement a DSPy-based pipeline in which each prediction includes a generated rationale and a citation to a governing rule, allowing extraction errors to be aggregated by attributed rule. This diagnostic signal is used by a Rationale-Oriented (RO) optimizer to perform targeted, human-readable signature revisions. Experiments use a representative 1000-sentence pool drawn from the CoNLL-2012 test split, with a nested 200-sentence development subset used for optimization and the remaining 800 sentences reserved as an unseen subset. After signature selection, the role-specific signatures are fixed; on the disjoint unseen subset, they improve over their corresponding unoptimized baselines for all nine evaluated 5W + 1H-aligned roles, although the magnitude of improvement and the development-to-held-out gap vary substantially by role, with the sparse ARGM-PRP and ARGM-CAU roles generalizing markedly less well than their development scores suggest. Across the full 1000-sentence pool, the selected-role, gold-predicate aggregate reaches 67.04% strict micro-F1; this is a within-study summary and not a full CoNLL SRL score. In the held-out ARG0 analysis, extraction outcomes are non-uniformly associated with cited rules, while signature perturbations show that rule citations are better interpreted as observable diagnostic proxies than as faithful per-instance causal explanations. Compared against MIPROv2 on the same extraction model, RO reaches a comparably low commission-error operating point without the recall collapse MIPROv2 incurs. Two further ablations locate the source of the effect: a full nine-role cross-model transfer to Phi-4-reasoning shows RO signatures carrying across model families unevenly, with gains concentrated on the weak, sparse roles and negligible-to-slightly-negative where the target model is already competent, and with transferred precision capped by its coarser span-boundary behavior; and a matched same-model self-critique ablation, in which Qwen3-14B supplies its own diagnostic feedback across the same nine roles, tests how much of the refinement effect can be recovered without a stronger external critic. Seen together, these ablations separate cross-model signature portability from teacher-capacity effects and clarify which parts of RO depend on the diagnostic structure itself versus the critic used to revise the signatures. The primary condition retains a design-time dependence on a proprietary teacher, and signature optimization is centered on a single open-weight extraction model, Qwen3-14B.
Global attention combines long-range lexical access with semantic-relation retrieval, making whole-layer replacement lossy. We present Strata-HeadQuotient, which statically assigns each key–value (KV) head to full-history attention (GLOBAL), 1024-token attention (LOCAL), or local attention plus an event-scoped typed graph read (LOCAL_GRAPH). For a 24-layer, 16-head decoder, we audited 384 heads and all 6903 pairs among 118 candidates before constrained subset selection. The Q25 export localized 96/384 heads: 81 were LOCAL and 15 were LOCAL_GRAPH. On 470 document-disjoint 8192-token windows (3.85 million tokens), graph-disabled Q25 perplexity was 1.00146× dense (95% document-bootstrap interval: 1.00131–1.00161); the 4096–8191-token ratio was 1.00201. On an untouched 2000-case semantic test, typed execution was 1998/2000, versus 374/2000 for a capacity-matched untyped graph. Two fresh adaptation/selection runs reproduced both directions, although every matched-capacity marginal graph-effect interval included zero. Q25 retained 78.125% of token-KV bytes and 98.44% of dense full-sequence throughput; a tested Q30 export failed. Thus, interaction-aware localization preserved 8192-token language-model quality, while a separate typed path controlled event-role behavior. The experiment does not establish that graph computation caused perplexity preservation.
Physics structure-informed neural networks (Ψ-NN) promise to carry known physical relations into compact models, but tiny machine learning (TinyML) deployment adds compression, finite precision, compilation, and hardware constraints that can change how those benefits appear in practice. We study this end to end for Burgers odd-in-x antisymmetry. A Ψ-NN preserves the encoded relation through clustering, 8-bit integer (INT8) quantization, compilation, and Cortex-M target-instruction-set emulation, achieving 6.0% relative L2 error and an antisymmetry residual of 0.006; none of seven evaluated compression baselines matches both under the common protocol. Deployment accounting reveals a complementary result. Centroid count is not deployable memory because the relation/index table must also be stored, limiting 32-bit floating-point (FP32) weight-memory reduction to about 4× under one-byte indexing. At matched parameter count, structured and dense models have comparable flash and instruction requirements, so structure improves Burgers accuracy and antisymmetry at a given footprint rather than automatically shrinking hardware requirements. Under complete lossless checkpointing, interruption timing has no independent effect, while budget loss and checkpoint degradation govern robustness. In the eleven-problem dataset, none of four tested descriptors reaches statistical significance for budget sensitivity. These results separate what physics structure preserves from what TinyML deployment must measure explicitly.
Recent studies have proposed post-processing strategies that iteratively reconstruct a partial segment of current solutions using Neural Combinatorial Optimization (NCO) solvers, thereby improving their performance on large-scale Vehicle Routing Problem (VRP) instances. However, the reconstruction subproblem is essentially a Shortest Hamiltonian Path Problem (SHPP) instance, which differs fundamentally from the original VRP variant on which the solver is trained. Consequently, the NCO solver may suffer from performance degradation during reconstruction due to limited generalization across problem variants. Moreover, relying solely on the solver may be insufficient to effectively identify and correct complex intersections or suboptimal topological structures. To address these limitations, we propose a post-processing strategy termed Hybrid Reconstruction Framework (HRC). Specifically, HRC first fine-tunes the NCO solver on SHPP instances and then exploits the enhanced solver to perform large-neighborhood random reconstruction. Subsequently, HRC conducts small-neighborhood reconstruction using 2-opt and kNN-DGR. The experimental results on both synthetic and real-world Traveling Salesman Problem and Capacitated Vehicle Routing Problem instances demonstrate that HRC substantially improves the performance of two representative NCO solvers on large-scale instances and achieves better overall performance than the state-of-the-art reconstruction strategy. Finally, ablation studies further validate the effectiveness of all proposed designs.
Remaining useful life (RUL) estimation is a fundamental task in Prognostics and Health Management (PHM), supporting condition-based and predictive maintenance of engineering systems. Data-driven methods contribute to many effective strategies for RUL prediction. However, two problems need to be solved when they are used in industrial applications: (1) The amount of data under one working condition is limited, and data from different working conditions suffer from domain discrepancies. These methods are constrained by distribution differences in data under different working conditions. (2) There is an urgent need to quickly achieve prediction with much less computing resources. To address these issues, a lightweight RUL prediction method called a knowledge transfer-based heterogeneous distillation network is proposed by combining knowledge distillation and transfer learning. First, the adversarial training mechanism is introduced for the extraction of domain-invariant features. Subsequently, a heterogeneous knowledge distillation framework is further designed for remaining useful life prediction, in which a bi-directional long short-term memory model serves as the teacher network and a compact fully connected network acts as the student model. The teacher model is used to learn informative degradation patterns and guide the training of the lightweight student model through knowledge transfer. Results obtained on the N-CMAPSS dataset verify that the proposed method achieves promising effectiveness and strong generalizability, reducing the average RMSE and MAE by 44.83% and 41.30%, respectively.
This study conducts a clinical evaluation of a secure, locally deployed, quantized large language model (LLM) for automating modified Rankin Scale (mRS) score extraction from unstructured neurosurgical notes. We retrospectively selected 103 authentic clinical letters (2007–2025) from aneurysm patients at a tertiary neurosurgical centre. To comply with data privacy constraints, an open-source reasoning LLM (Qwen3-32B) with 4-bit quantization was deployed entirely on-premises. The LLM extracted mRS scores using a zero-shot approach with custom logits processors to enforce strict JSON formatting. Performance was compared to a reference standard (attending neurosurgeons’ consensus) and parallel scoring by medical residents and students. The LLM achieved excellent agreement with the attending consensus (QWK 0.95), matching the reliability of medical students (QWK 0.95) and residents (QWK 0.93). Exact agreement was 75%, and agreement within ±1 mRS point was 96%. Bayesian analysis strongly supported statistical equivalence between the model and human raters. The computationally optimized LLM demonstrated human-level classification reliability without task-specific fine-tuning. This approach successfully addresses key patient data privacy barriers and the formatting inconsistencies typical of open-ended generative models. Securely deploying a general-purpose, quantized LLM provides a scalable pathway for extracting functional outcomes and supports FAIR-aligned data systems.
Client selection is a critical mechanism for ensuring robust convergence in Federated Learning (FL) systems, yet it remains vulnerable to Non-IID data distributions and Byzantine attacks. Deep Reinforcement Learning (DRL) has shown promise for automated client selection, yet existing methods suffer from three structural deficiencies: observation ambiguity, where scalar states cannot distinguish malicious updates from benign heterogeneity; reward fragility, whereby attackers exploit unbounded feedback to hijack policy updates; and risk blindness, as risk-neutral agents overlook the inherent variance in client contributions. To address these deficiencies, we propose InfRA-FL, a robust adaptive framework. A mutual-information-based state construction extracts high-utility features, resolving observation ambiguity. A Saturation-Aware Robust Reward (SARR) mechanism applies soft-clipping to bound each client’s influence on the policy gradient, provably neutralizing reward poisoning. URA-PPO, an uncertainty-aware algorithm with a dual-head Critic, optimizes a risk-penalized objective that shifts the agent from risk-neutral to risk-averse decision-making. Experiments on MNIST and CIFAR-10 under 20% Byzantine adversaries show that InfRA-FL outperforms state-of-the-art baselines by 5–10% in accuracy while accelerating convergence, establishing that principled information-theoretic observation and robust reward design suffice to secure RL-driven federated learning against targeted poisoning.
Target-speaker automatic speech recognition (TS-ASR) enables transcription of a specific speaker in multi-talker environments, yet remains largely unexplored for multilingual, low-resource languages. Existing TS-ASR systems predominantly target monolingual English using diarization-based or speaker-embedding approaches, leaving a critical gap for languages such as Kazakh, where code-switching with Russian and English is commonplace. We propose Persona-ASR, a modular two-stage architecture. The first stage is an explicit target-presence gate that verifies whether the enrolled speaker appears in the mixture and emits a token to suppress transcription when the speaker is absent, directly addressing the acoustic-hallucination failure mode of prior systems. The second stage performs enrollment-conditioned recognition: a 192-dimensional ECAPA-TDNN speaker embedding modulates a WavLM-Base-Plus encoder through feature-wise linear modulation (FiLM), while language-specific CTC heads enable joint Kazakh and English decoding without forcing Latin and Cyrillic symbols to compete in a single output space. To evaluate the system, we introduce KazMix3, a Kazakh overlap dataset for TS-ASR training, and PersonaMix, a controlled bilingual benchmark spanning same- and cross-language enrollment across varying interferer counts (1–3) and signal-to-noise ratios (−3 to +3 dB). Persona-ASR outperforms a strong off-the-shelf cascade baseline by 13.3 WER points on English and 24.6 on Kazakh, and matches a published monolingual English baseline. On PersonaMix, speaker conditioning reduces relative word error rate by 40.7% on English and 59.3% on Kazakh mixtures over an unconditioned variant of the same model, and cross-language enrollment (unseen during training) remains effective, increasing average raw WER by only 4.1 points (English) and 2.2 points (Kazakh) relative to same-language enrollment. To our knowledge, Persona-ASR is the first TS-ASR system for the Kazakh language, establishing a foundation for multilingual personalized ASR in low-resource settings.
The increasing digitization of healthcare records and growing reliance on interconnected systems have increased hospitals’ exposure to cyber threats, including ransomware, brute-force attacks, and Structured Query Language (SQL) injection. Traditional cybersecurity approaches alone are often insufficient to address these threats, prompting growing interest in artificial intelligence (AI)-based decision-support tools. This paper evaluates the potential of ChatGPT for hospital cybersecurity and incident response while introducing a structured qualitative framework for evaluating Large Language Model (LLM)-generated cybersecurity recommendations in healthcare. Through three progressively designed experiments and a ransomware case study, we evaluate ChatGPT’s role in developing a hospital cybersecurity plan, detecting brute-force login attempts, responding to an SQL injection attack, and managing a ransomware incident. Responses are assessed using five evaluation dimensions: specificity, completeness, technical correctness, feasibility, and alignment with the National Institute of Standards and Technology (NIST) Cybersecurity Framework (CSF), including both explicit mapping and function coverage. The results demonstrate that ChatGPT provides structured, context-aware guidance that aligns well with NIST CSF 2.0 and addresses governance and third-party risks. However, the recommendations also exhibit limitations, including limited operational depth, assumptions about technology and regulatory environments, lack of prioritization for resource-constrained settings, and limited consideration of implementation costs. Overall, the proposed evaluation framework provides a systematic approach for assessing LLM-generated cybersecurity guidance, while the findings indicate that ChatGPT can serve as a valuable decision-support tool that should complement, rather than replace, qualified cybersecurity professionals.
Quantum machine learning (QML) has emerged as a promising approach for analyzing the complex, high-dimensional data encountered in oncology, yet research in this area remains fragmented. This systematic literature review synthesizes current applications of QML in cancer care. Following PRISMA guidelines, six databases were searched for peer-reviewed English-language studies published between 2020 and 2026. Of the 212 records identified, 49 studies met the inclusion criteria after screening and quality assessment. The findings show that QML research is dominated by classification and detection tasks, while segmentation is beginning to emerge. Breast cancer and brain tumors are the most frequently investigated domains. Hybrid quantum-classical models, particularly quantum kernel methods, quantum neural networks, and quantum convolutional neural networks, are the predominant approaches. The main barriers to adoption are hardware limitations, including quantum noise, limited qubit availability, and reliance on simulators. Overall, QML in oncology remains in its early stages of development, with limited clinical validation, insufficient model interpretability, and little evidence of a clear quantum advantage. Future research should prioritize evaluation on real quantum hardware, larger and more diverse clinical datasets, standardized benchmarking against classical methods, and closer collaboration between computer scientists and oncology experts to facilitate clinical translation.
Autonomous agricultural robots require accurate trajectory tracking to perform precision field operations such as seeding, fertilization, and pruning. Classical kinematic models fail to capture the nonlinear dynamics inherent to real field conditions, limiting the performance of model-based control strategies. This work proposes an optimal hybrid methodology integrating Takagi–Sugeno (T–S) fuzzy model identification and Pure Pursuit (PP) control within a Particle Swarm Optimization (PSO) framework for a simulated pruning tractor. Data-driven T–S fuzzy models for incremental displacements MΔx and MΔy are identified using Fuzzy C-Means and parameterized via PSO. These fuzzy models are embedded in a PP feedback control scheme with discrete-time PI velocity and PD steering controllers, whose four gains are tuned by a second PSO instance. The fuzzy models achieve identification Root-Mean-Square Errors (RMSEs) of 10.598 × 10−3 m and 8.125 × 10−3 m. Integrated into the control loop, the system yields a lateral RMSE of 6.6 × 10−3 m on the training path and generalizes effectively across twelve complex agricultural coverage trajectories, maintaining a lateral RMSE below 12 × 10−3 m and heading RMSE under 1 degree. This interpretable, fuzzy rule-based approach provides an accurate and replicable simulation baseline for future experimental implementation on physical platforms.
This research introduces a new spectral pruning approach using the Adjacency Edge Index (AEI), which is a centrality measure from the theory of spectral graphs and first-order matrix perturbation theory. The AEI score reflects the contribution of each neuron to the network’s dynamic synchronizability via the Fiedler vector. The AEI approach thus offers a mathematically motivated saliency score in the context of data-driven neuron co-activation graphs. The proposed approach has been tested on MNIST, Fashion MNIST, KMNIST, and a real-world social network dataset. The approach has been extended to convolutional filter pruning on the CIFAR-10 dataset using spatial global average pooling. The AEI approach has been compared to magnitude-based pruning methods like L1 and L2 norms and gradient-based pruning methods like SNIP and GraSP. The robustness of the proposed approach has been established by comparing the results over five random seeds. The AEI approach is proposed as a principled, interpretable, structure-aware pruning criterion rather than an accuracy-maximising method. Spectral analysis demonstrates that AEI is the only evaluated method that systematically targets structurally peripheral neurons, whereas magnitude-based methods prune broadly across the structural spectrum and GraSP actively removes structurally central neurons, an effect most pronounced on more complex datasets and deeper architectures (CIFAR-100, ResNet-20), where it causes substantial accuracy degradation at high sparsity. The AEI approach has been extended to the Hybrid approach by combining the AEI and L2 norms. The Hybrid approach has been seen to improve the accuracy gap at 40% sparsity from 5.46 to 2.26 percentage points over the state of the art. The robustness of the proposed approach has been established by conducting ablation studies on the robustness of the approach to the selection of the graph. The approach has been seen to be moderately robust to the selection of the graph with Spearman’s ρ ≈ 0.70. The accuracy gap has been seen to be less than 0.2%.
This paper presents a novel method for pruning deep neural networks based on the concept of flow, derived from the continuous modeling of signal propagation across layers. We derive flow functions for fully connected, convolutional, and self-attention architectures, and we propose a new iterative pruning algorithm, Iterative Flow-Aware Pruning (IFAP), that leverages these measures to identify and eliminate non-essential parameters while preserving critical information pathways. Extensive experiments across ten prominent architectures (including CNNs, vision transformers, and efficient mobile networks) on ten benchmark datasets demonstrate consistent accuracy–compression trade-offs: 81% of the evaluated configurations achieve a 60–81% reduction in computational cost relative to the corresponding baseline model. Furthermore, 97% of the evaluated configurations retain more than 98% of their baseline Top-1 accuracy. These results validate flow-based importance scoring as a robust and general-purpose foundation for model optimization.
Background: Artificial intelligence (AI) and machine-learning (ML) prognostic models are increasingly developed for inflammatory bowel disease (IBD), yet their reported performance and clinical readiness remain inadequately appraised. Methods: Following PRISMA 2020 and a registered protocol, we searched PubMed, Web of Science, IEEE Xplore, and arXiv (January 2012–January 2026) for studies developing or validating prognostic models in Crohn’s disease or ulcerative colitis. Two reviewers independently screened the studies, extracted data, and assessed risk of bias using PROBAST + AI; discrimination was summarized by area under the curve (AUC) and stratified by validation type. Results: Of the 3050 records, 111 studies were included. Treatment response was the most common target; laboratory data and electronic health records were the most frequent modalities. Across 83 studies, the median AUC was 0.850; externally validated models reached 0.870 versus 0.845 for internal-only and 0.790 for cross-validation-only. External validation was reported in 29.7%, calibration in 14.4% and analysis code in 3.6%; the analysis domain was the leading source of bias. Conclusions: The evidence base maps reported discrimination rather than demonstrated clinical readiness. Until calibration, decision-curve utility, and transportability are reported alongside external validation, clinical deployment remains premature.
Temperature–time profiles obtained through thermistor-based monitoring provide a rich but noise-sensitive source of information for estimating kinetic and thermal parameters in exothermic batch reactions. Conventional workflows typically combine deterministic smoothing with numerical differentiation, an approach that can amplify measurement noise and fail to propagate preprocessing uncertainty into the resulting reaction-rate and parameter estimates. To address these limitations, this study presents an uncertainty-aware Bayesian machine-learning framework that integrates scalable random-Fourier-feature Gaussian-process (RFF–GP) smoothing, analytical differentiation, temperature-derived apparent conversion, Bayesian parameter inference, posterior validation, predictive calibration, model comparison, ablation, sensitivity analysis, probabilistic benchmarking, simulation of thermal nonideality, and endpoint diagnostics. The framework was applied to 379,631 cleaned thermistor observations. The production RFF–GP achieved a validation root-mean-square error of 0.04805K, yielding a stable latent temperature trajectory and an uncertainty-aware estimate of dT/dt. On a smaller matched subset, exact Gaussian-process regression achieved the highest predictive accuracy and the best probabilistic scores, whereas the RFF–GP reduced central-processing-unit runtime by approximately 4.1-fold and remained applicable to the larger production fit. A Monte Carlo dropout neural comparator produced larger prediction errors and substantially wider predictive intervals. Six apparent thermokinetic structures were evaluated using mean-field variational inference, after which the nth-order and autocatalytic structures were validated using the No-U-Turn Sampler (NUTS). Under mean-field variational inference, the apparent autocatalytic structure achieved the lowest point estimate of the widely applicable information criterion (WAIC), the lowest derivative-domain error, and the lowest full-profile temperature-reconstruction root-mean-square error of 0.2920K. Its posterior obtained using NUTS yielded Ea=40.98kJmol−1, kref=0.005815min−1, ΔTad=56.11K, m=0.1694, and n=1.0784. The sampling diagnostics indicated satisfactory convergence, large effective sample sizes, and no divergent transitions. Although the MFVI posterior means and NUTS posterior medians were similar, variational inference produced narrower uncertainty intervals for several correlated parameters. Moving-block bootstrap intervals did not establish a decisive separation in WAIC among the leading structures. Expanded sensitivity, ablation, imperfect-insulation simulation, and endpoint-holdout analyses further showed that the apparent parameter estimates were sensitive to optimization, thermal nonideality, sensor response, and Gaussian-process boundary behavior. The autocatalytic formulation should therefore be interpreted as the best-performing apparent structure among the candidates tested rather than as evidence of a unique chemical mechanism. Overall, the framework extracted physically plausible apparent thermokinetic information from noisy temperature-only measurements while explicitly quantifying uncertainty arising from prediction, parameter estimation, model form, computation, thermal nonideality, and boundary behavior.
Selective algorithmic auditing requires deciding which process evidence should receive attention when exhaustive review is infeasible. This Review introduces a four-layer framework that connects process representation, learning, inspection allocation, and governance within a single budgeted sequential decision problem over event streams. Unlike prior reviews centered on predictive process monitoring, explainability, cost analysis, or bibliometric structure, the proposed framework examines how these functions interact when human review, computation, latency, and documentation capacity are constrained. A structured and targeted survey of 89 unique publication families is used to illustrate and critically examine event-log, Petri-net, graph, object-centric, neural, uncertainty-aware, sequential, bandit, reinforcement learning, and audit architecture approaches. The reviewed evidence indicates that substantial bodies of work address the individual layers, but cross-layer evaluation remains fragmented and uses heterogeneous datasets, objectives, and validation protocols. The synthesis identifies five priorities: audit-ready benchmarks, explicit inspection budget protocols, calibrated uncertainty, transfer across organizational contexts, and reproducible governance interfaces. The main contribution is a computational framework and a corpus-bounded research agenda that connects representation, learning, inspection allocation, and governance for selective algorithmic auditing.
Automated detection of epileptic seizures from electroencephalogram (EEG) recordings is essential for timely clinical intervention and long-term patient monitoring. Deep learning achieves high accuracy, but its limited interpretability and computational demands restrict deployment in resource-constrained, real-time clinical environments; furthermore, classical machine learning studies have concentrated on a narrow group of well-known temporal features. This study systematically introduces and evaluates 25 less-explored time-domain features, 13 of which have no documented precedent as classification features in scalp EEG seizure detection, against 25 classical features and their 50-feature combination. Raw, unfiltered recordings from the CHB-MIT database were segmented into 10-second windows, and seven classical classifiers were optimized with GridSearchCV under subject-wise StratifiedGroupKFold cross-validation, with the data partitioned at the patient level such that no patient appeared in both the training and test sets. The multilayer perceptron trained on the combined 50-feature set performed best (accuracy 86.59%, F1-score 86.35%), exceeding the classical features alone by 5.1 and 4.6 percentage points, respectively; the less-explored features alone remained competitive (84.63%, 84.67%). SHAP analysis identified the exponent of the detrended fluctuation analysis (DFA)—a long-range, nonlinear measure of temporal correlation—as the most influential predictor. The curated feature set, its rigorous subject-level validation, and its interpretability provide a reproducible and computationally efficient foundation for future clinical deployment.
Detecting a few hundred fraudulent transactions among hundreds of thousands is an extreme class-imbalance problem where one miss can cost a full transaction value. Stacking heterogeneous classifiers is the standard recipe, yet under a leakage-free, precision–recall evaluation, its ranking gain over the best single model is inconsistent: sizable when the base learners are diverse, and negligible when they are redundant. The ensemble’s dependable value lies elsewhere: member disagreement is a usable, threshold-free estimate of epistemic uncertainty. We propose Daunt (Disagreement As UNcertainty for Triage), which turns this disagreement into three deployment layers over one ensemble: (i) routing the most uncertain transactions to human review by ranking them on base-learner disagreement, (ii) deciding alarms by an example-dependent rule that weighs the fraud score against the transaction amount, and (iii) attaching interpretations verified for faithfulness and stability. On two contrasting datasets, the anonymized ULB (0.17% fraud) and feature-rich IEEE-CIS (3.5% fraud), deferring the 5% most uncertain transactions raises system recall from 0.76 to 0.91 (ULB) and 0.66 to 0.75 (IEEE-CIS) while removing every automated false alarm, and the example-dependent rule recovers more fraudulent money than any global threshold. Daunt reframes the heterogeneous ensemble as a source of actionable uncertainty rather than an end in itself.
Given the emergence of new applications with stringent networking requirements, traditional bit-level communication may reach theoretical Shannon capacity and struggle to support such applications. Therefore, the idea of semantic communication (SC) has been proposed in literature. SC involves the use of artificial intelligence and a shared knowledge base to send a representation of data and reconstruct it at the receiver. In this paper, a SC system for document transmission is proposed. The system uses a transformer and a context encoder module to preserve the document’s knowledge and context information by quantifying the relationship between a current sentence and the preceding sentences. The semantic information is sent over a noisy channel, and a mutual information maximization model is used to reduce the effect of noise on the transmitted signal. The proposed system is trained in two stages; the first stage involves training the sentence-level transformers while the second stage involves training the sentence-level and the context-level transformers. The performance of the system is evaluated using two publicly available datasets. The results show that the performance of the proposed system is better than a sentence-level SC model in terms of BLEU score and sentence similarity over a Rayleigh fading channel.
A parking space system is a vital component of a smart transport management system, and it helps in managing parking spaces, reducing traffic congestion, and improving mobility. Deep learning helps in releasing smart parking systems. Despite their higher computational efficiency, lightweight convolutional neural networks (CNNs) may have limited feature representation capabilities and suffer from the vanishing gradient problem, which can reduce classification performance under challenging parking scenarios. To address this limitation, lightweight CNN models, such as MobileNetV2, EfficientNetB0, DenseNet, and ConvNeXt, are considered, with squeeze-and-excitation (SE) blocks incorporated to enhance channel-wise feature recalibration and stabilize gradients throughout the network. MobileNetV2 captures discriminative local features. EfficientNetB0 learns multi-scale semantic representations through compound scaling. DenseNet promotes hierarchical feature reuse. ConvNeXt extracts robust contextual features. The extracted deep features are subsequently reduced in dimensionality using Uniform Manifold Approximation and Projection (UMAP) while preserving the underlying manifold structure and removing redundant information. The reduced features are then classified by the ensembling of XGBoost, LightGBM, and support vector machine (SVM) classifiers, and their predictions are combined through a weighted stacking ensemble to exploit their complementary strengths and improve generalization. The proposed framework was experimentally evaluated on the PKLot, CNRPark and CNRPark + Ext datasets, achieving classification accuracies of 99.3%, 98.7% and 99.1%, respectively, demonstrating its effectiveness and robustness for real-world smart-city parking applications.
