Search Results (94)

Cooperative multi-UAV path planning under dynamic and adversarial conditions demands simultaneous satisfaction of safety, efficiency, and coordination constraints, yet existing swarm-intelligence and RL–swarm hybrids rely on deterministic switching rules, tabular states, and ad hoc training schedules. This paper proposes RL-JSO, a hybrid framework in which a dueling double deep Q-network with prioritized experience replay adaptively selects among the drift, passive, and active phases of a jellyfish search optimizer, replacing the deterministic time-control rule with a learned policy. The framework integrates a five-layer hierarchical safety control mechanism, a mastery-gated nine-stage curriculum, and a shared reward module that architecturally enforces fairness between RL-JSO and a paired RL-PSO counterpart. Evaluation across four progressive campaigns with 160 independent runs per algorithm shows that, within the evaluated JSO/PSO family, RL-JSO is the only method that sustains a 100% collision-free rate across all four progressive difficulty campaigns, its Cliff’s delta over standard JSO grows monotonically with difficulty from medium to large, and under a composite cooperation metric its coordination score remains nearly invariant while comparators degrade by 17–23%. A paired inference-time ablation on the trained checkpoint provides controlled inference-time evidence that adaptive phase switching is a principal contributor to the observed test-time performance within the trained system, rather than the heuristic fallback layers. Full article

Foundation segmentation models exhibit strong generalization on natural images yet degrade substantially in underwater scenes due to color distortion, scattering, and low contrast, which collectively impair feature representation. Parameter-efficient fine-tuning strategies have been explored to adapt SAM to marine domains while preserving generalization, but degraded image quality still hampers feature extraction. Moreover, existing SAM-based underwater methods typically rely on ground-truth box prompts during inference. Since ground-truth boxes are inherently unavailable in real-world underwater scenarios, this dependence yields evaluation outcomes that fail to reflect actual deployment conditions, thereby limiting their practical applicability. To address these issues, Water-AutoSAM is introduced—a dual-domain enhanced auto-prompting framework tailored for underwater image segmentation. The auto-prompting mechanism decouples semantic and positional representations for generalized point generation, which are optimized via enhanced sharpness, correctness, and diversity losses under staged training. To counter the degrading effects typical of underwater imagery, a lightweight module designated SS-UIE is integrated as a frozen pre-enhancement stage. This module operates with spatial–frequency dual-branch processing and utilizes a fixed residual fusion coefficient to combine the two streams. Operating entirely without box prompts, Water-AutoSAM achieves competitive annotation-free performance, attaining 92.38% mIoU on SUIM and reducing the gap to the fully supervised upper bound to 2.08% on COD10K. Full article

Empathetic dialogue systems built upon large language models overwhelmingly adopt a monolithic inference paradigm that processes emotion perception, causal reasoning, memory retrieval, and response planning within a single forward pass without architecturally enforced intermediate representations, forfeiting intermediate-state transparency and long-horizon personalization. Drawing on neuroscientific and cognitive–psychological evidence that human empathy is functionally dissociable, we present MOSAIC (Multi-agent Orchestration with Structured Affective memory for Interpretable empathiC dialogue), a training-free framework that operationalizes empathetic dialogue as a four-stage cognitive pipeline: affective perception, causal appraisal, episodic memory retrieval, and response synthesis. Three innovations distinguish MOSAIC from prior work: (1) a cognitively motivated modular architecture whose functionally dissociable stages enable post hoc failure attribution through logged intermediate states; (2) a hierarchical three-tier emotional memory—perceptual, semantic, and episodic—coupled with adaptive three-dimensional retrieval over emotion, situation, and coping-strategy cues; and (3) a heterogeneous model orchestration strategy coordinating open-source and API-accessible models through role-specific chain-of-thought prompts, requiring no task-specific fine-tuning. We note that the EmpatheticDialogues evaluation pre-populates the memory store with 200 training-split episodes prior to test-set interaction, a data-access asymmetry relative to single-model baselines that must be borne in mind when interpreting comparative results. Experiments on EmpatheticDialogues and ESConv show that MOSAIC achieves a 76.4% weighted F1 and an empathy score of 3.87 (on a 1–5 Likert scale) and that it improves over single-model, training-free baselines on aggregate empathy and—most prominently—on human-rated personalization (3.67 vs. 3.24 against Claude-3.5 five-shot, $d=0.48$). We caution that the comparison against training-free baselines is not data access-controlled (see the cold-start discussion in Methods); the personalization advantage, supported by the ablation without the Event Agent, is the result we treat as the primary practical contribution of this work. Full article

A distinct typing interface using surface electromyography (sEMG) can facilitate silent, hands-free typing by interpreting muscle activity in relation to specific keystrokes. Character-level recognition poses greater challenges than coarse gesture recognition because it is sensitive to subtle temporal variations and overlapping muscle dynamics. Temporal features are essential for typing recognition because keypresses may differ in duration, force, and accompanying hand movements across users. This paper proposes TransTCNet, a two-stage deep neural network architecture with a causal convolutional layer for learning local features and a transformer-based component for learning long-range temporal interactions. We evaluated our network on a publicly available 26-class typing sEMG dataset acquired from 19 individuals. The model achieved a validation accuracy of 96.53%, exceeding the baseline models. Our study revealed generalization among participants, and the AUC values were also high (>0.994) across all classes. The model was highly reliable and exhibited high prediction confidence (>0.9), enabling us to achieve a high training accuracy (97.86%) for real-time filtering decisions. TransTCNet could be suitable for wearable and edge devices due to its efficient architecture and low inference cost. The model’s ability to consistently decode fine-grained neuromuscular signals across users makes it well-suited for real-time applications such as adaptive user interfaces, virtual and augmented reality, prosthetic control, and communication systems. Full article

Large Vision–Language Models (LVLMs) have achieved strong performance in multimodal understanding and generation. However, they remain prone to hallucination, where generated content deviates from the visual input, reducing output reliability. We analyze the attention mechanism and identify two key issues in visual information use. The model exhibits insufficient overall attention to visual tokens and weak or dispersed attention to semantically relevant regions, limiting effective visual grounding. We propose a tuning-free attention intervention method applied at inference time. In the encoding stage, we apply a structured rescaling to the attention logits associated with visual tokens, introducing a structural bias in the visual subspace. In the decoding stage, we filter attention heads based on their response magnitudes and perform weighted aggregation using their global response intensities. This design reinforces salient visual evidence while suppressing weak or diffuse attention patterns. Experiments on CHAIR and POPE show that our method reduces hallucination without additional training. On the CHAIR benchmark, it reduces the sentence-level metric by 15.5% and the instance-level metric by 5.7% on average, while consistently improving performance across multiple LVLMs and maintaining strong results on general multimodal benchmarks such as MME. Full article

We propose a field-transformation-based framework for generating phase-only light-field holograms from a single RGB image. The method establishes an explicit pipeline from monocular scene inference to holographic wavefront synthesis, without requiring multi-view capture or task-specific hologram-network training. First, we construct a layered occlusion RGB-D model from the input image using monocular depth estimation, connectivity-based layer decomposition, and occlusion-aware inpainting, which provides a lightweight 3D prior for sparse-view rendering in the small-parallax regime. Second, we transform the rendered sparse RGB-D light field into a target complex wavefront on the recording plane through local frequency mapping, thereby bridging explicit scene geometry and wave-optical field construction. Third, we optimize the phase-only hologram under multi-plane amplitude constraints using a geometrically consistent initial phase and an error-driven adaptive depth-sampling strategy, which improves convergence stability and reconstruction quality under a limited computational budget. Numerical experiments show that the proposed method achieves better depth continuity, occlusion fidelity, and lower speckle noise than representative layer-based and point-based methods, and improves the average PSNR and SSIM by approximately 3 dB and 0.15, respectively, over Hogel-Free Holography. Optical experiments further confirm the physical feasibility and robustness of the proposed framework. Full article

Maritime Domain Awareness (MDA) relies heavily on data acquired from high-resolution optical spaceborne sensors; however, processing this massive quantity of sensor data via traditional supervised deep learning is severely bottlenecked by its dependency on exhaustively annotated datasets. Under extreme data scarcity, conventional architectures suffer severe performance degradation, rendering them impractical for time-critical, zero-day deployments. To overcome this barrier, we propose a training-free inference paradigm that leverages the extensive pre-trained knowledge of Large Vision-Language Models (VLMs). Specifically, we introduce a Domain Knowledge-Enhanced In-Context Learning (DK-ICL) framework coupled with a Macro-Topological Chain-of-Thought (MT-CoT) strategy. This approach bridges the perspective gap between natural images and top–down optical sensor imagery by translating expert remote sensing heuristics into a strict, step-by-step reasoning pipeline. Extensive evaluations demonstrate the substantial efficacy of this framework. Armed with merely 4 visual exemplars per category as in-context triggers, our MT-CoT augmented VLMs outperform traditional models trained under identical scarcity by over 38% in F1-score. Crucially, real-world case studies confirm that this zero-gradient approach maintains robust generalization on unannotated, out-of-distribution coastal clutters, achieving performance parity with data-heavy networks trained on 50 times the data volume. By substituting massive human annotation and GPU optimization with scalable logical deduction, this paradigm establishes a resource-efficient foundation for next-generation intelligent maritime sensing networks. Full article

Full fine-tuning of large-scale vision-language models for video action recognition incurs prohibitive computational cost and often degrades pre-trained spatial representations. To address this, we propose VETA-CLIP, a Video Efficient Temporal Adaptation framework that enhances temporal modeling while preserving cross-modal alignment. By incorporating lightweight adapters into a frozen backbone, VETA-CLIP introduces only 3.55M trainable parameters (a 98% reduction compared to full fine-tuning). Our approach features two key innovations: (1) an Efficient Spatio-Temporal Attention (ESTA) mechanism with a parameter-free boundary replication temporal shift (BRTS) module, which explicitly decouples spatial and temporal attention heads to capture inter-frame dynamics while minimizing disruption to the pre-trained spatial representations; and (2) a novel Variation Loss that maximizes both local inter-frame differences and global temporal variance, encouraging the model to focus on action-related changes rather than static backgrounds. Extensive experiments on HMDB-51, UCF-101, and Something-Something v2 demonstrate that VETA-CLIP achieves competitive performance across zero-shot, base-to-novel, and few-shot protocols, while and remains competitive on the Kinetics-400 dataset. Notably, our eight-frame variant requires only 4.7 GB of peak GPU memory and 2.47 ms of inference per video, demonstrating exceptional computational efficiency alongside consistent accuracy gains. Full article

Introduction: Persons diagnosed with Autism Spectrum Disorder (ASD) require adaptations to dental care that many undergraduate programmes may not explicitly treat. This cross-sectional pilot study assessed the extent of ASD-related content in Australia and New Zealand (ANZ) dental and oral health curricula and explored Oral Health Therapy students’ knowledge and self-efficacy. Methods: Online surveys of academic staff across ANZ programmes and Bachelor of Oral Health Therapy students at the University of Newcastle were conducted. Quantitative data was summarised descriptively, and free text responses underwent thematic analysis. Results: Fifteen educator responses (8% of 178 invitees) suggest limited ASD-specific teaching and minimal use of simulation-based education. Among 38 student responses (from one institution), knowledge was generally foundational, but misconceptions persisted and no respondents reported high confidence in providing oral health care for Autistic patients. Interest in further training was high. Conclusions: Within the constraints of low response rates and a single institution student sample, these preliminary findings suggest opportunities to strengthen Autism-related teaching, particularly sensory adaptations, communication strategies, and experiential learning. Inferences should be considered exploratory and hypothesis generating. Limitations: Low educator responses and potential response bias due to limited external validity from a single student cohort. Full article

The arbitrary-order hidden Markov model ($\alpha$-HMM) is a nontrivial generalization of the standard HMM, designed to model stochastic processes with higher-order dependences among arbitrarily distant random events. The $\alpha$-HMM admits an efficient Viterbi-style optimal decoding algorithm, making it feasible to discover higher-order dependences among data objects in observed sequential data. Because the $\alpha$-HMM exceeds the expressive power of standard HMMs, fixed $k\mathrm{th}$-order HMMs, and stochastic context-free grammars, effective probabilistic parameter estimation approaches are required to translate this theoretical expressiveness of the $\alpha$-HMM into practical utility. This paper introduces a principled methodology for effective estimation of probabilistic parameters of the $\alpha$-HMM from observed data. In large-scale sequential datasets, higher-order dependencies can vary widely across instances, so a single global parameter set may be inadequate. Instead, an amortized parameter inference approach is proposed for the $\alpha$-HMM, in which an input-conditioned parameter estimator is learned from data and used to infer instance-specific parameters for each input instance to the decoding algorithm. Specifically, the neural parameter estimator is trained using a composite learning objective that is partially enabled by the optimal decoding algorithm. The effectiveness of the proposed parameter estimation method is demonstrated through empirical results of the application of the $\alpha$-HMM in biomolecular structure modeling and prediction. Full article

Accurate estimation of the mean-reversion speed $\alpha$ in the AR(1) process $X_{t+1}=(1-\alpha )X_t+{\epsilon }_t$ is central to energy-commodity modelling. Classical estimators such as GARCH, jump-diffusion, and regime-switching produce model-conditioned estimates by embedding $\alpha$ within distributional assumptions, so that different model choices yield different $\widehat{\alpha }$ values from the same series without a principled criterion to adjudicate. We propose a distribution-free neural estimator based on a Temporal Convolutional Network (TCN) trained on synthetic AR(1) series with Sinh-ArcSinh (SAS) innovations. Distribution-free here means that no parametric family is assumed for the innovation distribution at inference time: the estimator imposes no distributional hypothesis when processing a new series. The SAS family serves as a training vehicle—not a model for the real data—chosen for its ability to span a broad range of tail weights and asymmetry profiles. The theoretical foundation is spectral invariance: the Yule–Walker equations establish that the autocorrelation structure ${\rho }_k={(1-\alpha )}^k$ depends on $\alpha$ alone, provided innovations are uncorrelated across lags—a condition satisfied not only by i.i.d. innovations but also by conditionally heteroscedastic processes such as GARCH. The TCN therefore generalises to volatility-clustering environments without modification, learning to extract $\alpha$ from temporal dependence alone, independently of the marginal innovation distribution and of the temporal variance structure. On held-out test series the estimator outperforms all classical competitors, with the advantage growing monotonically with non-Gaussianity. A robustness analysis on three out-of-distribution innovation families and on AR(1)-GARCH(1,1) processes empirically validates the spectral invariance guarantee across both marginal and temporal variance structure, including near-integrated GARCH processes where innovation kurtosis far exceeds the training range. The distribution-free $\widehat{\alpha }$ enables a two-stage pipeline in which $\alpha$ and the innovation distribution are characterised independently—a decoupling structurally impossible in classical likelihood-based approaches. Once trained, the TCN acts as a universal mean-reversion estimator applicable to any price series without re-fitting. Applied to four energy markets—Italian natural gas (PSV price), Italian electricity (PUN price), US Henry Hub, and US PJM West Hub—spanning log-return kurtosis from near-Gaussian to strongly heavy-tailed, the TCN yields robust, distribution-free estimates of mean-reversion speed. Full article

Traffic flow forecasting provides essential support for the construction of smart transportation systems. Despite the superiority of the ASTGCN, which uses an attention mechanism to capture spatio-temporal correlations, it lacks an explicit physical interpretation and thus falls into a more general category known for its lack of such interpretation. As a result, in the presence of sparse or unstable congestion, these data-driven models often violate conservation laws and may generate “physical anomalies” or other logically impossible states. To close the gap of data-driven expressiveness and physical consistency, we propose the congestion-aware physics-guided STGCN (CAP-STGCN). This framework builds a synergistic model that achieves intrinsic coupling between the macroscopic traffic flow kinematics (fundamental diagram) and the spatio-temporal learning process. That is to say, under the model’s solution-space constraining effect, its motion space is bound on a feasible manifold. In terms of kinematics, it restricts consistency in the flow, density and speed. Concurrently, to address slow convergence under long-tailed distributions due to a lack of training samples, such as when there are fewer users or higher-quality items, a dynamic congestion-rectification mechanism is introduced. The aforementioned mechanism redefines the optimization landscape by prioritizing hard-to-predict saturation occurrences. Experiments show that, compared with other models, CAP-STGCN achieves higher prediction accuracy; more importantly, it is free of physical anomalies during inference and can be directly used in practice. Full article

We propose Adaptive Inference, a portfolio management framework extending Active Inference to non-stationary financial environments. The framework integrates inference, control, and execution under endogenous uncertainty, modeling investment decisions as coupled dynamics of belief updating, preference encoding, and action selection rather than optimization over fixed objectives. In this approach, portfolio behavior is governed by the expected free energy (EFE) minimization, showing that classical valuation models emerge as limiting cases when epistemic components vanish. Using train–test evaluation on the ARKK Innovation ETF (2015–2025), we identify a Passivity Paradox: frozen belief transfer outperforms naive adaptive learning. A Professional Agent achieves a Sharpe ratio of 0.39 while its adaptive counterpart degrades to $-0.28$, reflecting belief contamination when learning from policy-dependent signals. Crucially, the architecture is not designed to generate alpha but to perform endogenous risk management that mitigates overtrading under regime ambiguity and distributional shift. Adaptive Inference Agents maintain long exposure most of the time while tactically reducing positions during high-entropy periods, implementing uncertainty-aware passive investing. All agents reduce realized volatility relative to ARKK Buy-and-Hold (43.0% annualized). Cross-asset validation on the S&P 500 ETF (SPY) shows that inference-guided risk shaping achieves a positive Entropic Sharpe Ratio (ESR), defined as excess return per unit of informational work, thereby quantifying the economic value of information under thermodynamic constraints on inference. Full article

Adaptive Neuro-Fuzzy Inference System (ANFIS)-based inverse kinematics (IK) is highly accurate for trained poses but often yields approximations for unseen inputs due to non-standardized training data. This research addresses these limitations through two novel contributions designed for any generic Degrees of Freedom (DoF) serial revolute robotic arm. First, A structured training methodology is introduced using workspace decomposition and cubic path planning. Instead of random sampling, the workspace is partitioned into cubic regions where 28 unique trajectories (12 edges, 12 face diagonals, four space diagonals) connect the eight vertices using cubic polynomial interpolation. This ensures physically consistent data mirroring real world point to point (PTP) movements. Even though validated on an ABB IRB-1200 robotic arm, this modular design is inherently scalable, allowing the local cubic expertise to be extended to cover the entire reachable workspace. Second, a two-stage hybrid IK framework is proposed, where an initial ANFIS approximation is refined via Jacobian-based iterative methods. Three Hybrid Frame works were evaluated, Framework-1 (ANFIS + Jacobian Gradient), Framework-2 (ANFIS + Jacobian Pseudoinverse/Newton–Raphson), and Framework-3 (ANFIS + Damped Least Squares). The results show that all three hybrid IK frameworks achieve reliable convergence, while the DLS-based hybrid provides the best trade-off between accuracy, convergence speed, and numerical stability. This generic, analytical free architecture provides a computationally efficient solution even in a hybrid scenario, bridging the gap between offline structured training and online, real-time refinement for digital twin synchronization and industrial automation. Full article

Deep neural networks (DNNs) dominate text classification but suffer from high computational costs and poor generalization in data-scarce or Out-of-Distribution (OOD) environments. Conversely, non-parametric methods like compression-based offer robustness but incur prohibitive inference latency due to the reliance on exhaustive pairwise comparisons. To bridge this gap, this study proposes EGTJ, a training-free framework that introduces a novel retrieval-augmented compression architecture. Unlike prior works that apply similarity metrics in isolation, EGTJ utilizes an inverted-index pre-filtering mechanism to dynamically constrain the comparison scope, effectively reducing algorithmic complexity from linear to constant time relative to the training set size. Furthermore, a tri-metric fusion strategy is introduced that integrates information-theoretic (gzip), lexical (TF-IDF), and structural (Jaccard) similarities to mitigate the inherent biases of individual metrics. Experimental results across five in-distribution and four OOD datasets demonstrate that EGTJ achieves superior accuracy over all baseline methods—notably outperforming BERT by over 30% in 5-shot OOD scenarios—while simultaneously slashing inference latency by orders of magnitude compared to standard compression-based approaches. These findings present EGTJ as a scalable, high-performance alternative for resource-constrained NLP, effectively solving the scalability bottleneck of non-parametric classification. Full article