Search

• Bayesian brain theory (BBT) mathematically formalizes the dynamic of information processing through belief encoding and perceptual inference. • Beliefs are represented as probability densities about the latent causes of sensory data encoded in neural networks. • Organization of beliefs into hierarchies allows for the representation of a multitude of spatiotemporal scales and causal relationships within a unified probability space. • Belief-updating occurs through synaptic plasticity, relying on the timing of neuronal spikes and the balance of glutamatergic excitation and inhibition. • Precision (inverse of the variance) reflect the signal-to-noise ratio of sensory signals and is encoded by monoaminergic neuromodulators. Bayesian brain theory, a computational framework grounded in the principles of Predictive Processing (PP), proposes a mechanistic account of how beliefs are formed and updated. This theory assumes that the brain encodes a generative model of its environment, made up of probabilistic beliefs organized in networks, from which it generates predictions about future sensory inputs. The difference between predictions and sensory signals produces prediction errors, which are used to update belief networks. In this article, we introduce the fundamental principles of Bayesian brain theory, and show how the brain dynamics of prediction are associated with the generation and evolution of beliefs. [ABSTRACT FROM AUTHOR]

[Display omitted] • A novel 4SQR code is proposed based on traditional QR codes. • We implement the 4SQR to generate and detect the readability of the system. • The proposed 4SQR can increase data storing capacity with security. • It is tested in a digital twin framework using randomly generated samples. The usage of Quick Response (QR) Codes has become widely popular in recent years, primarily for immense electronic transactions and industry uses. The structural flexibility of QR Code architecture opens many more possibilities for researchers in the domain of the Industrial Internet of Things (IIoT). However, the limited storage capacity of the traditional QR Codes still fails to stretch the data capacity limits. The researchers of this domain have already introduced different kinds of techniques, including data hiding, multiplexing, data compression, color QR Codes, and so on. However, the research on increasing the data storage capacity of the QR Codes is very limited and still operational. The main objective of this work is to increase the data storage capacity of QR Codes in the IIoT domain. In the first part, we have introduced a 4-State-Pattern-based encoding technique to generate the proposed 4-State QR (4SQR) Code where actual data are encoded into a 4SQR Code image which increases the data storage capacity more than the traditional 2-State QR Code. The proposed 4SQR Code consists of four types of patterns, including Black Square Box (BSB), White Square Box (WSB), Triangle, and Circle, whereas the traditional 2-State QR Codes consist of BSB and WSB. In the second part, the 4SQR Code decoding module has been introduced using the adaptive YOLO V5 algorithm where the proposed 4SQR Code image is decoded into the actual data. The proposed model is tested in a Digital Twin (DT) framework using randomly generated 3000 testing samples for the encoding module that converts into 4SQR Code images successfully and similarly for the decoding module that decodes the 4SQR Code images into the actual data. Experimental results show that this proposed technique offers increased data storage capacity two times than traditional 2-State QR Codes. [ABSTRACT FROM AUTHOR]

This editorial sketches some general ideas and issues in the field of speech act syntax, and it highlights the development of the so-called cartographic approach to syntax in that domain. Given the recent focus of encoding speaker, addressee, and further speech act categories in the left periphery of the clause, the editorial points out the relevance of the contributions to this Virtual Special Issue. • State of the art of speech act syntax. • Distinction between speech act and clause type. • Development of the cartographic program in syntax. [ABSTRACT FROM AUTHOR]

The present study examines whether structural and functional variability in medial temporal lobe (MTL) neocortical regions correlate with individual differences in episodic memory and longitudinal memory change in cognitively healthy older adults. To address this question, older adults were administered a battery of neuropsychological tests on three occasions: the second occasion one month after the first test session, and a third session three years later. Structural and functional MRI data were acquired between the first two sessions and included an in-scanner associative recognition procedure enabling estimation of MTL encoding and recollection fMRI BOLD effects. Encoding effects in parahippocampal cortex correlated with associative recognition performance and baseline cognitive ability. Recollection effects in entorhinal cortex correlated with associative recognition performance and predicted memory change over the three-year follow-up interval, an association that remained after controlling for chronological age and entorhinal cortical volume. These findings suggest that entorhinal recollection effects may be indicative of the future functional integrity of the region and, hence, its capacity to support future memory performance. • Entorhinal recollection effects correlated negatively with longitudinal memory change. • Entorhinal recollection effects correlated with associative memory performance. • Parahippocampal cortex encoding effects correlated with memory and cognitive ability. [ABSTRACT FROM AUTHOR]

Many fMRI papers in the mind-reading literature present decoders as a final product, arguing that decoders provide neurophysiological insight and have real-world utility. However, mental state decoders are commonly built in a way that precludes straightforward physiological interpretations. This undermines the claim that decoders are interpretable or capture 'representations' of mental states. In contrast to decoding models, encoding models of task fMRI are computationally straightforward and more interpretable. Mind-reading research would benefit from shifting its focus from successful decoding per se to understanding how decoding is affected by different experimental parameters, which would demonstrate the information that decoders are sensitive to (e.g., the color of a banana versus its orientation). Decoding mental and perceptual states using fMRI has become increasingly popular over the past two decades, with numerous highly-cited studies published in high-profile journals. Nevertheless, what have we learned from these decoders? In this opinion, we argue that fMRI-based decoders are not neurophysiologically informative and are not, and likely cannot be, applicable to real-world decision-making. The former point stems from the fact that decoding models cannot disentangle neural mechanisms from their epiphenomena. The latter point stems from both logical and ethical constraints. Constructing decoders requires precious time and resources that should instead be directed toward scientific endeavors more likely to yield meaningful scientific progress. [ABSTRACT FROM AUTHOR]

Modern language models and vector-symbolic architectures show that vector-based models are capable of handling the compositional, structured, and symbolic properties required for human concepts. Vectors are also able to handle key phenomena from the psychology, including computation of features and similarities, reasoning about relations and analogies, and representation of theories. Language models show how vector representation of word semantics and sentences can interface between concepts and language, as seen in definitional theories of concepts or ad hoc concepts. The idea of Church encoding, from logic, allows us to understand how meaning can arise in vector-based or symbolic systems. By combining these recent computational results with classic findings in psychology, vector-based models provide a compelling account of human conceptual representation. For decades, cognitive scientists have debated what kind of representation might characterize human concepts. Whatever the format of the representation, it must allow for the computation of varied properties, including similarities, features, categories, definitions, and relations. It must also support the development of theories, ad hoc categories, and knowledge of procedures. Here, we discuss why vector-based representations provide a compelling account that can meet all these needs while being plausibly encoded into neural architectures. This view has become especially promising with recent advances in both large language models and vector symbolic architectures. These innovations show how vectors can handle many properties traditionally thought to be out of reach for neural models, including compositionality, definitions, structures, and symbolic computational processes. [ABSTRACT FROM AUTHOR]

• Neonatal P1 latencies to speech sounds are associated with rapid serial naming speed at the age of 28 months. • Larger amplitudes and shorter latencies of P1 reflect stronger neural speech encoding and predict better pre-reading skills. • A reduced contribution of the left hemisphere in neural speech encoding is related to poorer pre-reading skills. We longitudinally investigated whether infant P1 and N2 ERPs recorded in newborns and at 28 months could predict pre-reading skills at 28 months and 4–5 years. We recorded ERPs to a pseudoword in newborns and at 28 months in a sample over-represented by infants with familial dyslexia risk. Using multiple linear regression models, we examined P1 and N2 associations with pre-reading skills at 28 months and 4–5 years. Shorter latencies of the newborn P1 predicted faster serial naming at 28 months. Larger amplitudes and shorter latencies of P1 at 28 months predicted better serial naming abilities and auditory working memory across the pre-reading stage. Right-lateralized P1 and N2 were related to poorer pre-reading skills. Infant ERPs, particularly P1, providing information about neural speech encoding abilities, are associated with pre-reading skill development. Infant and early childhood neural speech encoding abilities may work as early predictive markers of reading development and impairment. This study may help to plan early interventions targeting phonological processing to prevent or ameliorate learning deficits. [ABSTRACT FROM AUTHOR]

To provide personalized recommendations for users, recommendation systems must predict their unknown preferences. Latent factor models consistently achieve high prediction accuracy for this task. During training, these models encode users and items as latent vectors. Aligning these vectors facilitates predicting the item preference of the user. The training process optimizes the objective function, aiming to minimize the disparity between latent vectors and the collected preferences. In this study, we aim to integrate item descriptions into the construction of the objective function due to the sparsity and inaccuracy of collected preferences. This process is accomplished using Bert for vectorizing item descriptions. We conducted experiments with the proposed approach on datasets Movielens 1M and Yahoo Webscope R4. Our approach demonstrates a reduction in RMSE compared to previous approaches within this research domain. [ABSTRACT FROM AUTHOR]

In the Transformer architecture, positional encoding is a vital component because it provides the model with information about the structure and position of data. In Graph Transformer, there have been attempts to introduce different positional encodings and inject additional structural information. Therefore, in terms of integrating positional and structural information, we propose a Structural and Positional Ensembled Graph Transformer (SPEGT). We developed SPEGT by noting the different properties of structural and positional encodings of graphs and the similarity of their computational processes. We have set a unified component that integrates the functionalities: (i) Random Walk Positional Encoding, (ii) Shortest Path Distance between each node, and (iii) Hierarchical Cluster Encoding. We find a problem with a well-known positional encoding and experimentally verify that combining it with other encodings can solve their problem. In addition, SPEGT outperforms previous models on a variety of graph datasets. We also show that SPEGT using unified positional encoding, performs well on structurally indistinguishable graph data through error case analysis. [Display omitted] • We combined information of graph structures to create a novel positional encoding. • The components of the new unified encoding compensate for each other's shortcomings. • Our new architecture continuously injects unified information into the attention. • Our model performs better when edge information is scarce. • The model discriminates graphs that are structurally indistinguishable. [ABSTRACT FROM AUTHOR]

This study aims to know the determinants of demonstrative selection in contexts where speakers of Jordanian Arabic (JA) use deictic demonstratives without aid from distinguishing descriptions or gestures to individuate a referent that exists with a similar entity in the same communicative situation. Data collection depended on a prolonged observation of JA use in natural settings. Our findings show that when multiple similar entities are in a status of competing referents in the same context, the speaker's selection of a demonstrative to individuate a specific entity is primarily determined by perceptions of the degree of visual prominence concerning that entity in relation to the similar entity. Visual prominence in our data involved the domains of more/less visibility, foreground/background, front side/back side, vertical/horizontal position, and more/less motion. Within each of these domains, 'proximals', were generally associated with the modes of access of high visibility, foreground, vertical position, and more motion; the opposite modes were associated with 'distals'. As a cognitive consequence of the perceptions concerning the visual prominence of a referent in relation to these modes, entities that were encoded with 'proximals' can be considered as having high accessibility while those encoded with 'distals' can be considered as having low accessibility. • Visual prominence determines demonstrative selection involving competing referents. • Visual prominence involves the domains and modes of access concerning the referents. • Proximals and distals encode referents that belong to different modes of access. • The modes of access can be correlated with high and low referent accessibility. • Proximals encode referents that have high accessibility and vice versa for distals. [ABSTRACT FROM AUTHOR]

Sentiment classification and emotion recognition are two close related tasks in NLP. However, most of the recent studies have treated them as two separate tasks, where the shared knowledge are neglected. In this paper, we propose a multi-task interactive graph attention network with position encodings, termed MIP-GAT, to improve the performance of each task by simultaneously leveraging similarities and differences. The main proposal is a multi-interactive graph interaction layer where a syntactic dependency connection , a cross-task connection and position encodings are constructed and incorporated into a unified graphical structure. Empirical evaluation on two benchmarking datasets, i.e., CMU-MOSEI and GoEmotions, shows the effectiveness of the proposed model over state-of-the-art baselines with the margin of 0.18%, 0.67% for sentiment analysis, 1.77%, 0.89% for emotion recognition. In addition, we also explore the superiority and limitations of the proposed model. • Sentiment classification and emotion recognition are two correlative tasks in NLP. • But they are often treated as separative tasks. • In contrast to differences, commonalities of sentiment and emotion are often ignored. • Keys in the task: syntactic knowledge, cross-task interaction, position information. • Simultaneously incorporating the three key factors into a graph structure. • Performance boosts by the interaction of two tasks in a unified graphical structure. [ABSTRACT FROM AUTHOR]

Lossless Data compression is the process of reducing the size or the number of bits required to represent data, and Arithmetic coding is one of the popular lossless text compression techniques. This project focuses on lossless data compression using an innovative approach that combines arithmetic coding and Recurrent Neural Networks (RNN) with the help of encoder-decoder architecture. The proposed method aims to improve compression ratios compared to standard techniques. By leveraging the power of RNNs and arithmetic coding, encoder-decoder architecture effectively captures patterns and dependencies in the input data, enabling more efficient compression. The proposed approach achieved a remarkable improvement, achieving an impressive compression ratio that is, on average, 3.5 times smaller than that achieved by traditional methods. These results showcase the potential of developing improved general-purpose compressors based on neural networks and hybrid modeling. This demonstrates the potential of the proposed method for significantly enhancing lossless data compression performance. [ABSTRACT FROM AUTHOR]

The 6-Degree-of-Freedom (6-DoF) robotic grasping is a fundamental task in robot manipulation, aimed at detecting graspable points and corresponding parameters in a 3D space, i.e affordance learning, and then a robot executes grasp actions with the detected affordances. Existing research works on affordance learning predominantly focus on learning local features directly for each grid in a voxel scene or each point in a point cloud scene, subsequently filtering the most promising candidate for execution. Contrarily, cognitive models of grasping highlight the significance of global descriptors, such as size, shape, and orientation, in grasping. These global descriptors indicate a grasp path closely tied to actions. Inspired by this, we propose a novel bio-inspired neural network that explicitly incorporates global feature encoding. In particular, our method utilizes a Truncated Signed Distance Function (TSDF) as input, and employs the recently proposed Transformer model to encode the global features of a scene directly. With the effective global representation, we then use deconvolution modules to decode multiple local features to generate graspable candidates. In addition, to integrate global and local features, we propose using a skip-connection module to merge lower-layer global features with higher-layer local features. Our approach, when tested on a recently proposed pile and packed grasping dataset for a decluttering task, surpassed state-of-the-art local feature learning methods by approximately 5% in terms of success and declutter rates. We also evaluated its running time and generalization ability, further demonstrating its superiority. We deployed our model on a Franka Panda robot arm, with real-world results aligning well with simulation data. This underscores our approach's effectiveness for generalization and real-world applications. [ABSTRACT FROM AUTHOR]

Foreground segmentation algorithm aims to precisely separate moving objects from the background in various environments. However, the interference from darkness, dynamic background information, and camera jitter makes it still challenging to build a decent detection network. To solve these issues, a triplet CNN and Transposed Convolutional Neural Network (TCNN) are created by attaching a Features Pooling Module (FPM). TCNN process reduces the amount of multi-scale inputs to the network by fusing features into the Foreground Segmentation Network (FgSegNet) based FPM, which extracts multi-scale features from images and builds a strong feature pooling. Additionally, the up-sampling network is added to the proposed technique, which is used to up-sample the abstract image representation, so that its spatial dimensions match with the input image. The large context and long-range dependencies among pixels are acquired by TCNN and segmentation mask, in multiple scales using triplet CNN, to enhance the foreground segmentation of FgSegNet. The results, clearly show that FgSegNet surpasses other state-of-the-art algorithms on the CDnet2014 datasets, with an average F-Measure of 0.9804, precision of 0.9801, PWC as (0.0461), and recall as (0.9896). Moreover, the FgSegNet with up-sampling achieves the F-measure of 0.9804 which is higher when compared to the FgSegNet without up-sampling. [ABSTRACT FROM AUTHOR]

To verify the diagnostic performance of the loss of nigrosome-1 on susceptibility-weighted imaging (SWI) with compressed sensing-sensitivity encoding (CS-SENSE) and neuromelanin on neuromelanin-sensitive (NM) magnetic resonance imaging (MRI) for the diagnosis of Parkinson's disease (PD) and atypical Parkinsonism. A total of 195 patients who underwent MRI between October 2019 and February 2020, including SWI, with or without CS-SENSE, and NM-MRI, were reviewed retrospectively. Two neuroradiologists assessed the loss of nigrosome-1 on SWI and neuromelanin on the NM-MRI. The result of N-3-fluoropropyl-2-beta-carbomethoxy-3-beta-(4-iodophenyl) nortropane positron-emission tomography (PET) was set as the reference standard. When CS-SENSE was applied for nigrosome-1 imaging on SWI, the non-diagnostic scan rate was lowered significantly from 19.3% (17/88) to 5.6% (6/107; p =0.004). Diagnosis of PD and atypical Parkinsonism based on the loss of nigrosome-1 on SWI and based on NM-MRI showed good diagnostic value (area under the curve [AUC] 0.821, 95% confidence interval [CI] = 0.755–0.875: AUC 0.832, 95% CI = 0.771–0.882, respectively) with a substantial inter-reader agreement (κ = 0.791 and 0.681, respectively). Combined SWI and neuromelanin had a similar discriminatory ability (AUC 0.830, 95% CI = 0.770–0.880). Similarly, the diagnosis of PD was excellent. CS-SENSE may add value to the diagnostic capability of nigrosome-1 on SWI to reduce the nondiagnostic scan rates. Furthermore, loss of nigrosome-1 on SWI or volume loss of neuromelanin on NM-MRI may be helpful for diagnosing PD. • CS-SENSE on SWI for the nigrosome-1 reduced nondiagnostic scan and acquisition time. • CS-SENSE for the nigrosome-1 improved diagnostic performance for PD and Parkinsonism. • Neuromelanin on NM-MRI indicated a comparable diagnostic ability to nigrosome-1 on SWI. • NM-MRI had a lower nondiagnostic scan rate than SWI. [ABSTRACT FROM AUTHOR]

Maxwell or concomitant fields imprint additional phases on the transverse magnetization. This concomitant phase may cause severe image artifacts like signal voids or distort the quantitative parameters due to the induced intravoxel dephasing. In particular, double diffusion encoding (DDE) schemes with two pairs of bipolar diffusion-weighting gradients separated by a refocusing radiofrequency (RF) pulse are prone to concomitant field-induced artifacts. In this work, a method for reducing concomitant field effects in these DDE sequences based on additional oscillating gradients is presented. These oscillating gradient pulses obtained by constrained optimization were added to the original gradient waveforms. The modified sequences reduced the accumulated concomitant phase without significant changes in the original sequence characteristics. The proposed method was applied to a DDE acquisition scheme consisting of 60 pairs of diffusion wave vectors. For phantom as well as for in vivo experiments, a considerable increase in the signal-to-noise ratio (SNR) was obtained. For phantom measurements with a diffusion weighting of b = 2000 s/mm2 for each of the gradient pairs, an SNR increase of up to 40% was observed for a transversal slice that had a distance of 5 cm from the isocenter. For equivalent slice parameters, in vivo measurements in the brain of a healthy volunteer exhibited an increase in SNR of up to 35% for b = 750 s/mm2 for each weighting. These findings are supported by corresponding simulations, which also predict a positive effect on the SNR. In summary, the presented method leads to an SNR gain without additional RF refocusing pulses. • Proposes a novel method for concomitant field effect reduction for double diffusion encoding. • Demonstrates a positive effect on the SNR in phantom and in vivo. • Reveals importance for clinical high-amplitude gradient systems. [ABSTRACT FROM AUTHOR]

ETC2 is a widely-used texture compression format in Android devices and applications, so efficient ETC2 encoding can reduce application development time. We present QuickETC2-HQ, a set of improved ETC2 encoding techniques for real-time, high-quality texture compression. Our modifications to the luma-based approximations used in etcpak 1.0, the state-of-the-art encoder which integrated the QuickETC2 method, allow the execution of additional compression modes and more accurate error comparisons. As a result, the image quality of QuickETC2-HQ is improved compared to that of etcpak 1.0 and is comparable to that of the reference encoder, ETCPACK 2.74, with the fast mode. In terms of performance, QuickETC2-HQ is orders of magnitude faster than ETCPACK, making it practical for real-time application execution and offline production builds of applications. [Display omitted] • A set of improved ETC2 encoding techniques for real-time, high-quality texture compression. • Solving several quality issues appeared in real-time ETC2 encoding (e.g., block artifacts, blurring, posterization, etc.). • Orders of magnitude faster than the reference ETCPACK encoder. [ABSTRACT FROM AUTHOR]

Photochromism has shown to be a promising tool for improving the authenticity of commercially available products. Additionally, improving the engineering process of authentication patterns has been crucial to offer mechanically reliable anticounterfeiting nanofibers. Herein, the electrospinning technology was applied to develop mechanically reliable and photoluminescent silicon dioxide-based electrospun glass nanofibers (80–110 nm) embedded with lanthanide-activated aluminate (LA) nanoparticles (NPs; 1–2 nm) for anticounterfeiting purposes. The produced nanocomposite films exhibited photochromism from colorless in visible spectrum to green under ultraviolet irradiation. The nanofibrous film transparency was maintained by presenting the strontium aluminate pigment as nano-scaled particles, which improves its distribution and prevents the formation of aggregates in the electrospun glass nanofibrous bulk. After being excited at 365 nm, the nanofibers made of phosphor@glass (LANPs@GLS) displayed an emission band at 519 nm. Increases in the pigment ratio enhanced the hydrophobicity of the LANPs@GLS nanofibers without altering their intrinsic characteristics. The LANPs@GLS films exhibited fast and reversible photochromism without fatigue when activated by UV light. Transparency and flexibility were shown by the nanofibrous mats. The proposed technique is reliable for making a wide range of anticounterfeiting materials. [ABSTRACT FROM AUTHOR]

Neural material representations are becoming a popular way to represent materials for rendering. They are more expressive than analytic models and occupy less memory than tabulated BTFs. However, existing neural materials are immutable, meaning that their output for a certain query of UVs, camera, and light vector is fixed once they are trained. While this is practical when there is no need to edit the material, it can become very limiting when the fragment of the material used for training is too small or not tileable, which frequently happens when the material has been captured with a gonioreflectometer. In this paper, we propose a novel neural material representation which jointly tackles the problems of BTF compression, tiling, and extrapolation. At test time, our method uses a guidance image as input to condition the neural BTF to the structural features of this input image. Then, the neural BTF can be queried as a regular BTF using UVs, camera, and light vectors. Every component in our framework is purposefully designed to maximize BTF encoding quality at minimal parameter count and computational complexity, achieving competitive compression rates compared with previous work. We demonstrate the results of our method on a variety of synthetic and captured materials, showing its generality and capacity to learn to represent many optical properties. • Neural BTF representations allow for efficient reflectance encoding. • Autoencoder models trained with data augmentation allow for material propagation. • Neural reflectance propagation enables the creation of tileable BTF. [ABSTRACT FROM AUTHOR]

We propose a novel few-shot learning framework that can recognize human–object interaction (HOI) classes with a few labeled samples. We achieve this by leveraging a meta-learning paradigm where human–object interactions are embedded into compact features for similarity calculation. More specifically, spatial and temporal relationships of HOI in videos are constructed with transformers which boost the performance over the baseline significantly. First, we present a spatial encoder that extracts the spatial context and infers frame-level features of a human and objects in each frame. And then the video-level feature is obtained by encoding a series of frame-level feature vectors with a temporal encoder. Experiments on two datasets, CAD-120 and Something-Else, validate that our approach achieves 7.8% and 15.2% accuracy improvement on 1-shot task, 4.7% and 15.7% on 5-shot task, which outperforms the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

We used event-related potentials (ERPs) to examine encoding and retrieval during episodic memory in people with schizophrenia (SZ) and biological relatives of SZ (SZr). To isolate contextual from item-specific aspects of memory, we employed the Relational and Item-Specific Encoding (RISE) task. Twenty two healthy controls (HCs), 22 SZ, and 19 SZr, encoded visual depictions of objects when displayed alone (item-specific) or in pairs (relational encoding), and were later tested on recognition of specific objects and whether pairs of objects had appeared together. An early posterior component (P2) during encoding predicted later recognition and was diminished in SZ. A late negative potential (LNP) over left frontal brain regions during recognition was larger for relationally encoded objects than new and item-specific encoded objects in HCs. This pattern was absent for SZ and SZr. Smaller P2 and LNP components were associated with greater self-reported cognitive-perceptual abnormalities. Early posterior brain responses likely relevant to perceptual functions supporting memory formation were diminished in schizophrenia. Late frontal electrophysiological responses associated with relational aspects of memory appear diminished in SZ and SZr, potentially reflecting the influence of genetic liability for schizophrenia on brain functions supporting episodic memory. [ABSTRACT FROM AUTHOR]

Diffusion models have gained attention for their ability to represent complex distributions and incorporate uncertainty, making them ideal for robust predictions in the presence of noisy or incomplete data. In this study, we develop and enhance score-based diffusion models in field reconstruction tasks, where the goal is to estimate complete spatial fields from partial observations. We introduce a condition encoding approach to construct a tractable mapping between observed and unobserved regions using a learnable integration of sparse observations and interpolated fields as an inductive bias. With refined sensing representations and an unraveled temporal dimension, our method can handle arbitrary moving sensors and effectively reconstruct fields. Furthermore, we conduct a comprehensive benchmark of our approach against a deterministic interpolation-based method across various static and time-dependent PDEs. Our study attempts to addresses the gap in strong baselines for evaluating performance across varying sampling hyperparameters, noise levels, and conditioning methods. Our results show that diffusion models with cross-attention and the proposed conditional encoding generally outperform other methods under noisy conditions, although the deterministic method excels with noiseless data. Additionally, both the diffusion models and the deterministic method surpass the numerical approach in accuracy and computational cost for the steady problem. We also demonstrate the ability of the model to capture possible reconstructions and improve the accuracy of fused results in covariance-based correction tasks using ensemble sampling. [ABSTRACT FROM AUTHOR]

We present a novel approach for synthesizing realistic novel views using Neural Radiance Fields (NeRF) with uncontrolled photos. While NeRF has shown impressive results in controlled settings, it struggles with transient objects commonly found in dynamic and time-varying scenes. Our framework called Inpainting Enhanced NeRF , or IE-NeRF addresses this issue by learning to decompose the static and dynamic components of the scene through exploring transient mask inpainting. Specifically, to model varying appearances, we propose a CNN-based appearance encoder, and a view-consistent appearance loss to transfer consistent photometric appearance in different views. Moreover, our approach extends the Multi-Layer Perceptrons (MLP) network of NeRF, enabling it to simultaneously generate intrinsic properties (static color, density) and extrinsic transient masks. During optimization we take advantage of inpainting to restore static scene image while eliminating occlusions with the guidance of transient masks. Additionally, we propose a new training strategy using frequency regularization with transient mask factor in integrated positional encoding, which facilitates faster inference and early separation of transient components during training. We evaluate our approach on internet photo collections of landmarks, demonstrating its ability to generate high-quality novel views and outperform the existing methods. • The CNN-based appearance encoder for different views input. • Decomposition of the static and dynamic by exploring transient mask inpainting. • Frequency regularization base on transient mask factor in training. • Experiments and ablation studies confirm the effectiveness of our approach. [ABSTRACT FROM AUTHOR]

The inadequate awareness of real-world knowledge often causes machines to produce generic responses, such as 'I think so.', which may bring the degression of user interests. Consequently, enriching knowledge awareness and fabricating informative responses are long-standing challenges in open-domain dialogue systems. Previous studies have shown incorporating everyday commonsense knowledge can significantly enhance the open-domain dialogue response models. Nonetheless, previous works can only use explicit or implicit knowledge. Unlike them, this work presents a novel MixEI to leverage both explicit and implicit commonsense knowledge in dialogue generation. MixEI uses Dual-Way Knowledge Alignment and MixEI-Ranker to retrieve a set of contextually relevant commonsense facts as the explicit background knowledge and identify implicit knowledge labels by selecting clue facts that can tightly connect the dialogue context. MixEI uses BART as the backbone. After jointly encoding the background knowledge and dialogue history, MixEI first tries to externalize the implicit clue knowledge; then, the response decoding can seek information from both explicit and implicit knowledge. Extensive experiments on Chinese Weibo and English Reddit have verified the superior performance of the proposed MixEI-Ranker and MixEI. • A novel MixEI that can use both explicit and implicit commonsense knowledge. • A novel MixEI-Ranker that can efficiently retrieve relevant commonsense knowledge. • Extensive experiments have shown superior performance of our methods. [ABSTRACT FROM AUTHOR]

Mainstream backbone networks in computer vision, leveraging sophisticated CNN or Transformer architectures, have achieved remarkable performance across various tasks. However, these works primarily focus on downsample ranges of 4 × to 32 × on input images, with limited exploration into the embedding portion of the network. In this paper, we propose a novel Multi-Scale Embedding Block (MSEB) to enhance embedding features. MSEB is constructed through a combination of Self-Calibrated Channel Dilate Unit (SCDU) and Multi-Scale Encoder Unit (MSEU). SCDU projects images into high-dimensional space, incorporating channel self-calibration to preserve fine-grained information in shallow feature maps, thereby providing a more comprehensive representation for subsequent network layers. MSEU employs a multi-scale cascade design with implicit feature sharing characteristics, expanding the receptive field scale range to enhance embedding representation. Extensive experiments on mainstream benchmarks demonstrate that our proposed multi-scale embedding block can be seamlessly integrated into most popular CNN and Transformer architectures such as ResNet, Res2Net, Swin-Transformer, and MViT. By replacing the existing embedding structure in these methods, significant improvements on ImageNet-1k datasets are observed, e.g., CNN-based ResNet-50 and further adopting a multi-scale design Res2Net-50 achieve Top-1 accuracy improvements of 0.47% and 1.22%, respectively. Attention-based Swin-Tiny and MViT-Tiny, both adopting multi-scale design, achieved Top-1 accuracy improvements of 2.57% and 1.05%, respectively. • Too quick downsample during image embedding causes loss of image information. • A multi-scale embedding module is proposed to enhance the image embedding features. • A self-calibrating dilated unit is proposed to extract and preserve shallow feature. • A multi-scale encoding unit is proposed for extracting rich scale shallow features. [ABSTRACT FROM AUTHOR]

Span-based methods for nested named entity recognition (NER) are effective in handling the complexities of nested entities with hierarchical structures. However, these methods often overlook valid semantic dependencies among global spans, resulting in a partial loss of semantic information. To address this issue, we propose the Global Span Semantic Dependency Awareness and Filtering Network (GSSDAF). Our model begins with BERT for initial sentence encoding. Following this, a span semantic representation matrix is generated using a multi-head biaffine attention mechanism. We introduce the Global Span Dependency Awareness (GSDA) module to capture valid semantic dependencies among all spans, and the Local Span Dependency Enhancement (LSDE) module to selectively enhance key local dependencies. The enhanced span semantic representation matrix is then decoded to classify the spans. We evaluated our model on seven public datasets. Experimental results demonstrate that our model effectively handles nested NER, achieving higher F1 scores compared to baselines. Ablation experiments confirm the effectiveness of each module. Further analysis indicates that our model can learn valid semantic dependencies between global spans, significantly improving the accuracy of nested entity recognition. Our code is available at https://github.com/Shaun-Wong/GSSDAF. [ABSTRACT FROM AUTHOR]

Remarkable progress has been achieved in image generation with the introduction of generative models. However, precisely controlling the content in generated images remains a challenging task due to their fundamental training objective. This paper addresses this challenge by proposing a novel image generation framework explicitly designed to incorporate desired content in output images. The framework utilizes advanced encoding techniques, integrating subnetworks called content fusion and frequency encoding modules. The frequency encoding module first captures features and structures of reference images by exclusively focusing on selected frequency components. Subsequently, the content fusion module generates a content-guiding vector that encapsulates desired content features. During the image generation process, content-guiding vectors from real images are fused with projected noise vectors. This ensures the production of generated images that not only maintain consistent content from guiding images but also exhibit diverse stylistic variations. To validate the effectiveness of the proposed framework in preserving content attributes, extensive experiments are conducted on widely used benchmark datasets, including Flickr-Faces-High Quality, Animal Faces High Quality, and Large-scale Scene Understanding datasets. [ABSTRACT FROM AUTHOR]

This paper presents a 12-bit, 4.5 MS/s synchronous SAR ADC implemented in SMIC's 130-nm CMOS technology. It features a novel method for internal common-mode (V cm) generation via introduces switched-capacitor technique, eliminating the need for external V cm generation circuitry and reducing parasitics. To enhance capacitor matching and mitigate mismatch, thermometer encoding is employed for the most significant 6-bit of the ADC. The comparator boasts a dynamic pre-amplifier with switching capacitance tail current for improved signal amplification. The ADC exhibits excellent linearity with DNL of +0.59/−0.57 LSB and INL of +1.02/−1.12 LSB. Occupying a core area of 0.144 mm2, it achieves SNDR of 69.63 dB, ENOB of 11.27-bit, SFDR of 78.17 dB, and THD of −76.33 dB. Operating at 3.3 V, it consumes 2.577 mW with FoM W of 231.9 fJ/conversion-step. These measurement results confirm the feasibility and effectiveness of the proposed V cm self-generation approach, offering fresh insights into V cm generation in ADC design, with potential implications for future ADC architectures. [ABSTRACT FROM AUTHOR]

Episodic events are typically retrieved and forgotten holistically. If you recall one element (e.g., a person), you are more likely to recall other elements from the same event (e.g., the location), a pattern that is retained over time in the presence of forgetting. In contrast, representations of individual items, such as objects, may be less coherently bound, such that object features are forgotten at different rates and retrieval dependency decreases across delay. To test the theoretical prediction that forgetting qualitatively differs across levels in a representational hierarchy, we investigated the potential dissociation between event and item memory across five experiments. Participants encoded three-element events comprising images of famous people, locations, and objects. We measured retrieval accuracy and the dependency between the retrieval of event associations and object features, immediately after encoding and after various delays (5 h to 3 days). Across experiments, retrieval accuracy decreased for both events and objects over time, revealing forgetting. Retrieval dependency for event elements (i.e., people, locations, and objects) did not change over time, suggesting the holistic forgetting of events. Retrieval dependency for object features (i.e., state and colour) was more variable. Depending on encoding and delay conditions across the experiments, we observed both fragmentation and holistic forgetting of object features. Our results suggest that event representations remain coherent over time, whereas object representations can, but do not always, fragment. This provides support for our representational hierarchy framework of forgetting, however there are (still to be determined) boundary conditions in relation to the fragmentation of object representations. [ABSTRACT FROM AUTHOR]

Perceptual Video Compression (PVC) is one of the most promising approaches to enhancing compression efficiency. Saliency and Just Noticeable Distortion (JND) are the most common visual models in PVC. Most methods solely rely on one of them to construct PVC schemes, while a few methods attempt to integrate both. However, the latter are often too simplistic and coarse, failing to fully exploit them to reduce bitrate. To address this issue, we propose an efficient PVC scheme based on Deep Learning-Assisted Saliency and JND (SJ-PVC), which is implemented on the latest Versatile Video Coding (VVC). Specifically, we first design a Structurally Simplified Network (SS-Net) for video saliency prediction, which removes the redundancy of multi-scale models and maintains excellent accuracy. Then, an Adaptive Quantization Parameter (QP) selection algorithm based on Saliency and JND (SJAQP) is proposed. It can dynamically adjust QP offsets according to the characteristics of the encoding unit and effectively integrate the offset effects of saliency and JND. Finally, we design a Rate-Distortion Optimization based on Saliency and JND (SJRDO), which incorporates JND to obtain perceptual distortion terms and adjusts rate–distortion balance through saliency to achieve more rational bitrate allocation. This scheme fully considers the properties of both visual models and further eliminates visual redundancy. Extensive experiments demonstrate that, when maintaining subjective visual quality, SJ-PVC can save 22.87% of the bitrate compared to VVC while also shortening encoding time, significantly enhancing video coding efficiency. [Display omitted] [ABSTRACT FROM AUTHOR]

Text recognition, especially for complex scripts like Chinese, faces unique challenges due to its intricate character structures and vast vocabulary. Traditional one-hot encoding methods struggle with the representation of hierarchical radicals, recognition of Out-Of-Vocabulary (OOV) characters, and on-device deployment due to their computational intensity. To address these challenges, we propose HierCode, a novel and lightweight codebook that exploits the innate hierarchical nature of Chinese characters. HierCode employs a multi-hot encoding strategy, leveraging hierarchical binary tree encoding and prototype learning to create distinctive, informative representations for each character. This approach not only facilitates zero-shot recognition of OOV characters by utilizing shared radicals and structures but also excels in line-level recognition tasks by computing similarity with visual features, a notable advantage over existing methods. Extensive experiments across diverse benchmarks, including handwritten, scene, document, web, and ancient text, have showcased HierCode's superiority for both conventional and zero-shot Chinese character or text recognition, exhibiting state-of-the-art performance with significantly fewer parameters and lower floating-point operations (FLOPs). • HierCode represents Chinese characters using hierarchical encoding and prototype learning. • HierCode exhibits lightweight characteristics and lower FLOPs. • HierCode achieves SOTA results in zero-shot Chinese character and line recognition. [ABSTRACT FROM AUTHOR]

Current Transformer-based natural language understanding (NLU) models heavily rely on dataset biases, while failing to handle real-world out-of-distribution (OOD) instances. Many methods have been proposed to deal with this issue, but they ignore the fact that the features learned in different layers of Transformer-based NLU models are different. In this paper, we first conduct preliminary studies to obtain two conclusions: 1) both low- and high-layer sentence representations encode common biased features during training; 2) the low-layer sentence representations encode fewer unbiased features than the high-layer ones. Based on these conclusions, we propose a simple yet effective self-debiasing framework for Transformer-based NLU models. Concretely, we first stack a classifier on a selected low layer. Then, we introduce a residual connection that feeds the low-layer sentence representation to the top-layer classifier. In this way, the top-layer sentence representation will be trained to ignore the common biased features encoded by the low-layer sentence representation and focus on task-relevant unbiased features. During inference, we remove the residual connection and directly use the top-layer sentence representation to make predictions. Extensive experiments and in-depth analyses on NLU tasks demonstrate the superiority of our framework, achieving a new state-of-the-art (SOTA) on three OOD test sets. [ABSTRACT FROM AUTHOR]

For applying symbolic planning, there is the necessity of providing the specification of a symbolic action model, which is usually manually specified by a domain expert. However, such an encoding may be faulty due to either human errors or lack of domain knowledge. Therefore, learning the symbolic action model in an automated way has been widely adopted as an alternative to its manual specification. In this paper, we focus on the problem of learning action models offline, from an input set of partially observable plan traces. In particular, we propose an approach to: (i) augment the observability of a given plan trace by applying predefined logical rules; (ii) learn the preconditions and effects of each action in a plan trace from partial observations before and after the action execution. We formally prove that our approach learns action models with fundamental theoretical properties, not provided by other methods. We experimentally show that our approach outperforms a state-of-the-art method on a large set of existing benchmark domains. Furthermore, we compare the effectiveness of the learned action models for solving planning problems and show that the action models learned by our approach are much more effective w.r.t. a state-of-the-art method.1 [ABSTRACT FROM AUTHOR]

• Some neuron-like behaviors in a commercial VCSEL are theoretically and experimentally studied. • Image encoding and recovery based on the excited spiking patterns in a 1550 nm VCSEL is theoretically and experimentally realized, and the image recovery rate (IRA) between the recovered binary images and the original binarized images for Baboon and Lena images can be over 98 %. • A Post-processing technique for the spiking trains is proposed to recovery the binarized images. • The proposed image encoding and recovery method can provide a new path for the application of VCSEL neuron in future large-scale and complicated photonic spiking neural networks. We propose and investigate an image encoding and recovery scheme based on an excited 1550-nm vertical-cavity surface-emitting laser (VCSEL). The results show that VCSEL subjected to orthogonal optical injection can exhibit some neuron-like behaviors including injection locking, spiking exciting and inhibition, and damped oscillations. Based on the excited spiking patterns, the image can be encoded into continuous, up to GHz-rate optical spike trains, and then be successfully recovered after adopting a post-processing technique. Additionally, the experimental results are almost identical to that of theoretical simulations, and the image recovery accuracy (IRA) can reach 98.3% in experiment. This proposed image encoding and recovery approach can open a new window for future high-speed all-optical signal processing in the photonic spiking neural networks (PSNN). [ABSTRACT FROM AUTHOR]

• A novel snapshot full-Stokes polarizing-spectral-imaging framework is proposed by incorporating the advanced metasurface design for polarizing-spectrally dependent PSF engineering and the post computation. • The metasurface encodes the polarization-spectral imaging information through anisotropic detour phase modulation scheme, which is in principle suitable for a wide bandwidth and can overcome the restriction on wavelength selection. • The method shows advantages of compact configuration, fast acquisition and wideband operation. A broadband snapshot full-Stokes polarizing-spectral-imaging (PSI) is achieved by reciprocally designing a metal–insulator-metal metasurface with anisotropic detour phase to generate polarizing-spectrally dependent dot-array-shaped point-spread-function (PSF), and the high dimensional information can be retrieved from one snapshot intensity image through the post-computation. It overcomes the restriction on the wavelength selection in principle faced by the propagating phase modulation scheme used in the previous metasurface-based PSI approaches, and brings advantages including compact configuration, fast acquisition and broadband operation. The dependence of the engineered PSF on the polarizing-spectral channel across 100 nm in the visible band is experimentally verified and analyzed, consistent well with the simulation. Reconstruction results for the high dimensional imaging are presented in simulations and experiments based on the established model. [ABSTRACT FROM AUTHOR]

The existing key point detection network models have complex structures, so it is difficult to deploy on edge devices. Meanwhile, the convolution is localized and limited by the size of convolution kernels, which cannot effectively capture long-range dependencies. To address this problem, this paper introduces a lightweight Convolutional Transformer network (LHFormer Net) for human pose estimation. Considering that the sampling area of convolution kernels for different dimensional feature maps is fixed and the contextual information is singular, the enhanced receptive field block is designed to extract richer feature information and reduce information loss in feature maps. Based on the global modeling features of Transformer encoder, convolutional position encoding and multi-head self-attention are used to capture the spatial constraint relationship between key points in the deep feature extraction. Finally, a lightweight deconvolution module is used to generate higher resolution features to achieve multi-resolution supervision, which can effectively solve the problem of scale variation in pose estimation, more accurately locate key points of small and medium-sized people, ad further improve the detection accuracy of the network. Compared with other networks, the experimental results on the open access datasets COCO2017 and MPII show that the proposed network achieves a good balance between model complexity and detection performance. [ABSTRACT FROM AUTHOR]

Sparse large-scale multi-objective optimization problems (LSMOPs), in which most decision variables of the Pareto-optimal solutions are zero, have become increasingly prevalent in real-world applications. An effective approach for addressing the sparse LSMOPs is the two-layer encoding scheme. However, a significant challenge when adapting the two-layer encoding scheme remains in accurately identifying the sparse distribution of Pareto-optimal solutions. Therefore, this paper proposes a sparse large-scale multi-objective evolutionary algorithm based on sparsity detection. The proposed algorithm uses the two-layer encoding scheme with a specialized focus on finding the positions of sparse non-zero variables by optimizing the binary vector. In the algorithm, an initialization strategy based on sparsity detection is proposed. This strategy employs a sparsity detection method to identify non-zero variables at the beginning of the algorithm. Moreover, a two-stage search strategy is designed to enhance the ability to search the sparse Pareto-optimal solutions. In this strategy, two types of knowledge are adaptively selected to guide the genetic operators of the binary vector. Experimental results from the benchmark suite and real-world applications demonstrate that the proposed algorithm outperforms the existing state-of-the-art algorithms in solving sparse LSMOPs. [ABSTRACT FROM AUTHOR]

Modeling spatio-temporal sequences is an important topic yet challenging for existing neural networks. Most of the current spatio-temporal sequence prediction methods usually capture features separately in temporal and spatial dimensions or employ multiple mutually independent local spatio-temporal graphs to represent a spatio-temporal sequence. The first kind of method mentioned above is difficult to mine the complex spatio-temporal correlations, while the other is limited for the accuracy of long-term predictions. To handle these issues, this paper proposes a Transformer-Improved Graph Convolution Network for spatio-temporal prediction. Specifically, the temporal location encoding method is exploited to derive the spatio-temporal characteristics of the sequence utilizing a spatio-temporal feature fusion network. In addition, a spatio-temporal attention network is developed to enhance the spatio-temporal correlation of the sequence, and the dynamic spatial features of sequence are further extracted through the adaptive graph convolution network. A private dataset and a public dataset are employed to demonstrate the performance of the proposed TIGC-Net. The qualitative and quantitative results show that the proposed TIGC-Net can extract dynamic spatio-temporal properties more effectively, enhance the spatio-temporal correlation of sequences and improve the prediction accuracy compared with four state-of-the-art. [ABSTRACT FROM AUTHOR]

Accurate preoperative measurement of abdominal aortic aneurysms (AAAs) and associated vascular structures is indispensable for surgical planning. However, the precise extraction of complex multi-branch vascular structures from CTA images presents significant challenges. These challenges stem from the high individual variation in vascular structure, the difficulty of separating important organ branches from the surrounding tissue area, and the unclear boundaries between main branches and organ branches. To overcome these obstacles, the SDLU-Net network architecture is introduced. Firstly, a novel spatial positional encoding structure, named "dual position encoding", is presented to preserve the spatial relationships of vascular anatomical structures throughout the network operations. Secondly, a similarity-based dynamic attention linking module is proposed to effectively distinguish vessels from tissues, and the main branches from other branches. Furthermore, dual skip connections and channel attention modules are integrated into the decoding layer to enhance the information flow within the network. Lastly, a hybrid loss function was introduced to address the issue of sample imbalance, thereby improving the segmentation results of organ branches. Extensive experiments have demonstrated that SDLU-Net shows excellent segmentation performance in multi-branch blood vessels of the abdominal aorta, highlighting its substantial clinical potential for surgical planning involving complex multi-branch structures. • Dual positional encoding is presented to preserve the spatial relationships of vascular anatomical structures. • A similarity-based dynamic attention linking module to improve the segmentation accuracy. • Dual skip connections and channel attention modules are used in the decoder. • A hybrid loss function addresses the class imbalance. [ABSTRACT FROM AUTHOR]

Semi-supervised video action detection has received increasing attention due to its lower data annotation cost and performance comparable to fully supervised methods. However, due to the presence of dynamic background regions in the video, existing methods may encounter biases when interpreting the foreground and background of the video. This bias causes the model to mistakenly identify dynamic background areas as action foregrounds or to overlook background information, leading to misjudgment of the foreground. In response to this issue, this paper proposes a multi-granularity hierarchical contrastive learning between foreground and background for semi-supervised video action detection method termed as Multi-FB. Specifically, this paper proposes a multi-granularity encoding network based on foreground and background. This network uses a unified encoder to represent and learn foreground and background regions in videos at different granularities, thereby improving the model's understanding of action foreground and related background. Secondly, this paper proposes an Intra-model multi-granularity hierarchical contrastive strategy, which aims to minimize the representation discrepancies of foreground-to-foreground and background-to-background at different granularities within the same video, while maximizing the representation differences between the foreground and background at various granularities within the video. Furthermore, this paper proposes a Cross-model multi-granularity hierarchical contrastive strategy, which aims to enhance the consistency of the model's representations of foregrounds and backgrounds between the original data and the augmented data. A large number of experimental results on JHMDB-21 and UCF101–24 show that the proposed method can significantly distinguish feature representations between different categories, achieving performance comparable to state-of-the-art methods under semi-supervised conditions. [ABSTRACT FROM AUTHOR]

The input of the multi-document summarization task is usually long and has high redundancy. Encoding multiple documents is a challenge for the Seq2Seq architecture. The way of concatenating multiple documents into a sequence ignores the relation between documents. Attention-based Seq2Seq architectures have slightly improved the cross-document relation modeling for multi-document summarization. However, these methods ignore the relation between sentences, and there is little improvement that can be achieved through the attention mechanism alone. This paper proposes a hierarchical approach to leveraging the relation between words, sentences, and documents for abstractive multi-document summarization. Our model employs the Graph Convolutional Networks (GCN) for capturing the cross-document and cross-sentence relations. The GCN module can enrich semantic representations by generating high-level hidden features. Our model achieves significant improvement over the attention-based baseline, beating the Hierarchical Transformer by 3.4/1.64, 1.92/1.44 ROUGE-1/2 F1 points on the Multi-News and WikiSum datasets, respectively. Experimental results demonstrate that our delivered method brings substantial improvements over several strong baselines. [ABSTRACT FROM AUTHOR]

Topic models have become ubiquitous tools for analyzing streaming data. However, existing streaming topic models suffer from several limitations when applied to real-world data streams. This includes the inability to accommodate evolving vocabularies and control topic quality throughout the streaming process. In this paper, we propose a novel streaming topic modeling approach that dynamically adapts to the changing nature of data streams. Our method leverages Byte-Pair Encoding embedding (BPEmb) to resolve the out-of-vocabulary problem that arises with new words in the stream. Additionally, we introduce a topic change variable that provides fine-grained control over topics' parameter updates and present a preservation approach to retain high-coherence topics at each time step, helping preserve semantic quality. To further enhance model adaptability, our method allows dynamical adjustment of topic space size as needed. To the best of our knowledge, we are the first to address the expansion of vocabulary and maintain topic quality during the streaming process. Extensive experiments show the superior effectiveness of our method. [ABSTRACT FROM AUTHOR]

[Display omitted] • The paper presents a hormone sensor using a gold metasurface and graphene to detect reproductive hormones. • The sensor achieves a sensitivity of 375 GHz/RIU, an FOM of 7.693 RIU−1, and a detection limit of 0.556 RIU. • The sensor operates in the terahertz (THz) range, using unique molecular fingerprints for selective hormone detection. • The K-Nearest Neighbors (KNN) model enhances the sensor's accuracy, showing improved performance with optimized parameters. • The sensor supports two-bit encoding through graphene chemical potential modulation. This study presents an advanced sensor system integrating gold metasurfaces with graphene for the detection of reproductive hormones via refractive index variations. The proposed design operates under terahertz (THz) excitation to exploit the unique molecular signatures of hormones and minimize potential interferences. Comprehensive numerical analysis and finite element method (FEM) simulations were conducted to optimize the sensor's design parameters and evaluate its efficacy. The optimized sensor exhibits exceptional performance metrics, including a peak sensitivity of 375 GHzRIU−1, a figure of merit (FOM) of 7.693 RIU−1, a quality factor (Q) of 2.381, and a limit of detection (LOD) of 0.556 RIU. Furthermore, the sensor demonstrates two-bit encoding capabilities through modulation of graphene's chemical potential. Integration of a K-Nearest Neighbors (KNN) Regressor model enhances the sensor's accuracy while reducing required resources and simulation time by approximately 85 %. These findings demonstrate the sensor's potential for diverse applications, particularly in the biomedical field. [ABSTRACT FROM AUTHOR]

• It enables dual-frequency methods to achieve high accuracy with four stripes. • A reliable decoding strategy is proposed to enhance the noise immunity. • It encodes more high-frequency cycles without additional images. Dual-frequency fringe projection techniques have been broadly utilized for three-dimensional (3D) shape recovery. However, a pervasive challenge across these methodologies lies in the precise implementation of high-speed measurements. This paper explores a spatial coding strategy designed for the dual-frequency phase unwrapping method to tackle this. The strategy mandates the utilization of only four patterns, encapsulating the phase information of both high and low frequencies. The codewords are strategically embedded within the low-frequency phase, facilitating the acquisition of an accurate absolute phase map. A robust decoding algorithm engineered to ensure dependable phase unwrapping, particularly in scenarios characterized by a substantial number of high-frequency cycles. The proposed method exhibits superior noise immunity and enhances accuracy when compared with the previous dual-frequency phase-shifting methodology. Simulations and experiments demonstrate the robustness and efficiency of the proposed strategy. [ABSTRACT FROM AUTHOR]

[Display omitted] • The anti-counterfeiting SERS tag (Ag@SiO 2) encoding with flexible information was fabricated. • The anti-counterfeiting Ag@SiO 2 shown high security, excellent stability and anti-interference property. • The Ag@SiO 2 was successfully printing on surface of paper through security ink and shown excellent anti-counterfeiting effect. The core–shell structured Ag@SiO 2 nanomaterial integrated with surface-enhanced Raman scattering (SERS) spectroscopy promises a critical application in anti-counterfeiting. Security labels have been fabricated based on Ag@SiO 2 embedded with Raman reporters. The Ag@SiO 2 nanomaterial shows good stability and excellent anti-interference property for anti-counterfeiting. Multiple kinds of Raman probe molecules have been anchored in the Ag@SiO 2 labels to provide specific and abundant encoding information. The flexible encoding security information could be controlled conveniently by adjusting probe molecules, which not only enrich the SERS information but also improve the level of anti-counterfeiting. Furthermore, the Ag@SiO 2 shown excellent stability in organic solvent, and successfully used in ink for the anti-counterfeiting application. [ABSTRACT FROM AUTHOR]

The joint extraction of elements such as entities, relations, events, and their arguments, along with their specific interrelationships, is a critical task in natural language processing. Existing research often employs implicit handling of interactions between tasks through techniques like shared encodings or parameter sharing, lacking explicit modeling of the specific relationships between tasks. This limitation hinders the full utilization of inter-task correlation information and impacts effective collaboration between tasks. To address this, we propose a model for joint extraction of elements and relations based on causal relationship representation enhancement (CRE). This model aims to capture specific relationships between tasks in multiple stages, facilitating finer adjustments and optimization of subtasks and thereby enhancing the overall model performance. Specifically, CRE comprises three key modules: feature adaptation, feature interaction, and feature fusion. The feature adaptation module selects and adjusts features from shared encodings based on the requirements of specific tasks to better adapt to semantic differences between different tasks. The feature interaction module employs causal reasoning to comprehensively capture the causal relationships between tasks while mitigating negative transfer brought about by interference from semantic information. The feature fusion module further integrates features to obtain optimized task-specific representations. Ultimately, CRE exhibits a significant improvement in average performance across multiple information extraction tasks. [ABSTRACT FROM AUTHOR]

The multibeam sonar has become a powerful tool for undersea pipeline detection, which can obtain high quality three-dimensional point cloud data. However, the identification accuracy of current pipeline algorithms remains low because of insufficient utilization of context features. Therefore, this paper proposes an improved context-learning point cloud segmentation network, named Cross-scale Point-In-Context (Cross-PIC), to identify submarine pipelines. Cross-PIC designs the sampling module Balanced-FPS to improve the equalization sampling. Meanwhile, Cross-PIC constructs different scale coding architecture based on context learning, which includes Auxiliary line and Baseline for multi-scale feature encoding. Finally, the Refine Module is designed to realize cross-scale coded feature fusion to output segmentation results. Experiments with different fuzzy degrees show that Cross-PIC also significantly outperforms the original Point-In-Context network with less training data. Cross-PIC also achieves better results than other networks based on point, voxel, and dynamic graph convolution. • Single-scale conventional inspection methods are highly inaccurate for marine pipeline detection. • A proposed multi-scale feature fusion network architecture for multibeam point clouds based on in-context learning. • Equalization module is designed to solve the target imbalance problem in multibeam point clouds. • Refine module is designed to implement corrections to the results of submarine pipeline surveys. [ABSTRACT FROM AUTHOR]

The current image captioning directly encodes the detected target area and recognizes the objects in the image to correctly describe the image. However, it is unreliable to make full use of regional features because they cannot convey contextual information, such as the relationship between objects and the lack of object predicate level semantics. An effective model should contain multiple modes and explore their interactions to help understand the image. Therefore, we introduce the Multi-Modal Graph Aggregation Transformer (MMGAT), which uses the information of various image modes to fill this gap. It first represents an image as a graph consisting of three sub-graphs, depicting context grid, region, and semantic text modalities respectively. Then, we introduce three aggregators that guide message passing from one graph to another to exploit context in different modalities, so as to refine the features of nodes. The updated nodes have better features for image captioning. We show significant performance scores of 144.6% CIDEr on MS-COCO and 80.3% CIDEr on Flickr30k compared to state of the arts, and conduct a rigorous analysis to demonstrate the importance of each part of our design. [ABSTRACT FROM AUTHOR]

In a previous study from the author and his co-worker, a constitutive model for isotropic, elastic perfectly-plastic materials was developed using scalar, conjugate, stress/strain base pairs. These stress/strain base pairs result from a Gram–Schmidt decomposition of the deformation gradient. A limitation of this prior work is that we assumed the microscopic structures of such elastic–plastic materials would remain constant throughout a deformation process, i.e., there would be no change in the resulting microstructural properties. Typically, internal state variables are used to represent macroscopic manifestations of these microstructural properties. In this paper, we incorporate internal state variables into our previously developed constitutive model, and as a consequence, plastically-induced anisotropy shows up naturally in the developed model. These are, however, different from the typically used internal state variables and are introduced only for this purpose. • Plasticity-induced anisotropy is incorporated into a QR kinematics-based material model. • The anisotropy parameters, used in a encoding/decoding map, are considered as variables for this purpose. • A co-rotational rate of the internal state variables is defined and used in the model. • Relevant examples have been provided to demonstrate the utility of the proposed model. [ABSTRACT FROM AUTHOR]