Fine Granularity Research Articles

VHR-BirdPose: Vision Transformer-Based HRNet for Bird Pose Estimation with Attention Mechanism

Pose estimation plays a crucial role in recognizing and analyzing the postures, actions, and movements of humans and animals using computer vision and machine learning techniques. However, bird pose estimation encounters specific challenges, including bird diversity, posture variation, and the fine granularity of posture. To overcome these challenges, we propose VHR-BirdPose, a method that combines Vision Transformer (ViT) and Deep High-Resolution Network (HRNet) with an attention mechanism. VHR-BirdPose effectively extracts features using Vision Transformer’s self-attention mechanism, which captures global dependencies in the images and allows for better capturing of pose details and changes. The attention mechanism is employed to enhance the focus on bird keypoints, improving the accuracy of pose estimation. By combining HRNet with Vision Transformer, our model can extract multi-scale features while maintaining high-resolution details and incorporating richer semantic information through the attention mechanism. This integration of HRNet and Vision Transformer leverages the advantages of both models, resulting in accurate and robust bird pose estimation. We conducted extensive experiments on the Animal Kingdom dataset to evaluate the performance of VHR-BirdPose. The results demonstrate that our proposed method achieves state-of-the-art performance in bird pose estimation. VHR-BirdPose based on bird images is of great significance for the advancement of bird behaviors, ecological understanding, and the protection of bird populations.

MFGAD-INT: in-band network telemetry data-driven anomaly detection using multi-feature fusion graph deep learning

As the cloud services market grows, cloud management tools that detect network anomalies in a non-intrusive manner are critical to improve users’ experience of cloud services. However, some network anomalies, such as Microburst, in cloud systems are very discreet. Network monitoring methods, e.g., SNMP, Ping, are of coarse temporal granularity or low-dimension metrics, have difficulty to identify such anomalies quickly and accurately. Network telemetry is able to collect rich network metrics with fine temporal granularity, which can provide deep insight into network anomalies. However, the rich features in the telemetry data are insufficient exploited in existing research. This paper proposes a Multi-feature Fusion Graph Deep learning approach driven by the In-band Network Telemetry, shorted as MFGAD-INT, to efficiently extract and process the spatial-temporal correlation information in telemetry data and effectively identify the anomalies. The experimental results show that the accuracy performance of the proposed method improves about 10.56% compared to the anomaly detection method without network telemetry and about 9.73% compared to the network telemetry-based method.

E-commerce brand authenticity perception model of territorial characteristic agricultural products based on fuzzy cognitive map and emotional analysis

Brand authenticity perception is essential for territorial characteristic agricultural product e-commerce studies. From the complexity of consumer perception of brand authenticity, an e-commerce brand authenticity perception (EBAP) analysis model is proposed based on fuzzy cognitive map (FCM) and emotional analysis of online comments. Firstly, LDA model and snowNLP tools extract consumer perception attributes and their emotional inclination. After that, FCM and improved Bonferroni mean (BM) operator are used to accurately analyze the interrelationships between different attributes and comprehensively evaluate the brand authenticity of different enterprises under the same characteristic agricultural product. Finally, the model comparison experiment results show that the model proposed can reflect the “attribute importance” and “emotional inclination” of the e-commerce brand authenticity perception of territorial characteristic agricultural products. Among them, “platform logistics” and “product benefits” are essential in promoting the authenticity of brand-consumer relationships. Meanwhile, “e-commerce aftersales service” is closely related to the positive evaluation of “platform logistics” and “product benefits.” This study expands the methodical approach to brand authenticity perception research; it provides a valuable reference for developing modern fine granularity management of e-commerce brand authenticity for characteristic agricultural products.

Distributed intelligence for pervasive optical network telemetry

Optical network automation and failure management require measuring the status and the performance of the different network devices to anticipate any degradation and ensure the quality of the provided services, i.e., optical connectivity. Such pervasive network telemetry entails collecting large amounts of measurements and events from different sources and with very fine granularity, which given the amount and variety of telemetry sources and the size of each measurement and event, imposes requirements that are hard to achieve without large investments. In this paper, we analyze the main limitations of telemetry architectures relying exclusively on centralized systems for data analysis and propose an architecture with distributed intelligence. Data aggregation techniques, especially conceived for optical network telemetry, are presented with the objective of reducing data dimensionality. Illustrative results from our experimental telemetry system reveal a reduction of 3 orders of magnitude in terms of total data volume without introducing significant error and processing delay and, more importantly, helping network automation algorithms to identify meaningful changes in the network status.

Toward a Value-Based Therapy Recommendation Model.

Patient value is an important factor in clinical decision making, but conventionally, it is not incorporated in the decision processes. Clinical decision making has some clinical guidelines as a reference. There are very few value-based clinical guidelines, but knowledge about how values affect decision making is mentioned in some scattered studies in the literature. We use a literature review method to extract evidence and integrate it as part of the decision-making model. In this paper, a value-based therapy recommendation comprehensive model is proposed. A literature analysis is conducted to collect value-based evidence. The patients' values are defined and classified with fine granularity. Categorized values and candidate therapies are used in combination as filtering keywords to build this literature database. The literature analysis method generates a literature database used as a source of arguments for influencing decision making based on values. Then, a formalism model is put forward to integrate the value-based evidence with clinical evidence, and the literature databases and clinical guidelines are collected and analyzed to populate the evidence repository. During the decision-making processes, the evidence repository is utilized to match patients' clinical information and values. Decision-makers can dynamically adjust the relative importance of the two pieces of evidence to obtain a treatment plan that is more suitable for the patient. A prototype system was implemented using a case study for breast cancer and validated for feasibility and effectiveness through controlled experiments.

A novel granular ball computing-based fuzzy rough set for feature selection in label distribution learning

Label distribution learning is a widely studied supervised learning diagram that can handle the problem of label ambiguity. The increasing size of datasets is accompanied by the disaster of dimensionality, which implies that the arrival of redundant and noisy features undermines the effect of label distribution learning. As a crucial data-preprocessing technique, feature selection is capable of choosing discriminative features. However, due to the complex issue of label ambiguity, traditional feature selection approaches for datasets with logical labels cannot be applied to label distribution data. In this paper, a novel granular ball computing-based fuzzy rough set (GBFRS) is proposed for label distribution feature selection. Specifically, the proposed method is first introduced at the finest granularity, i.e., calculating similarity relations between single data points. Considering that the label ambiguity issue is exacerbated by the label imbalance phenomenon, the relative similarity in label distribution space among samples is computed for better generalization of the model. Then, a robust approximation strategy is devised for the target sample by using its true different and partially different class samples. Finally, with the concept of granular balls, the method explores the similarity relations between balls and samples, and the granular ball computing-based fuzzy rough set method is developed , which is endowed with the ability to simulate the characteristics of large-scale priorities in human thinking and considers local consistency. Extensive experiments conducted on twenty-two datasets show that GBFRS can effectively select more significant features than seven state-of-the-art feature selection algorithms.

Multi-stage affine motion estimation fast algorithm for versatile video coding using decision tree

Affine motion estimation (AME) which is newly introduced in versatile video coding (VVC) plays a significant role in the bit-rate reduction for rotation and zooming scenes. However, it brings the complexity extremely increased. In this paper, a fast AME algorithm is designed to cope with the high-complexity problem. Three crucial stages are involved in the fast algorithm, which is the optimal inter-mode decision based on the coding unit (CU) partition, the fast algorithm in affine motion search (AMS), and the fast inter-mode decision based on a decision tree. Specifically, the optimal inter-mode will be determined straightforwardly when the parent CU is checked whether it is translational motion estimation (TME). Then, three fast algorithms are made in AMS, which are early termination based on control point motion vector (CPMV), early termination of the iteration process, and fast fine granularity CPMV search. Finally, the optimal inter-mode is predicted using a decision tree based on eight essential features. Experimental results show that the proposed algorithm achieves 10.20% and 10.33% encoding time-saving on average under Low Delay B (LDB) and Random Access (RA) configuration, while the BD-Rate loss is only 0.12% and 0.14%, respectively.

Exploring intra-urban human mobility and daily activity patterns from the lens of dockless bike-sharing: A case study of Beijing, China

In recent years, the emergence of dockless bike-sharing has brought new ways of transportation, while also providing support for observing urban dynamics at a finer granularity. This study uses dockless bike-sharing data, combined with point of interest (POI) data, to explore intra-urban human mobility and daily activity patterns, using Beijing as a case study. Here, we employ spatial statistics and graph network to quantify the characteristics of travel behavior. Firstly, spatiotemporal statistics analysis was conducted to investigate the travel patterns and spatiotemporal differences of bike-sharing in Beijing. It was found that bike trips have clear peak hours and highly overlap with the subway lines, indicating that the bike-sharing system effectively completes the ‘last mile’ of travel in Beijing. Secondly, graph networks were used to reveal the core patterns and community structures of human mobility in Beijing and their connections to daily activities. Our findings indicate that there are different core structures of human mobility in Beijing, some dominated by a single core node, while others are formed by the aggregation of multiple cores. Finally, POI data was used to analyze human daily activity patterns. We analyzed the temporal characteristics and the hotspot areas of each daily activity in Beijing and found that transferring to other transportation modes is the most popular activity for people riding bikes, followed by shopping, working, and eating out. These findings provide a reference for understanding urban human dynamics and urban structure from the perspective of dockless bike-sharing.

Examining Psychologist Prescriptive Authority as a Cost-Effective Strategy for Reducing Suicide Rates.

Six states (NM, LA, IL, IA, ID, CO) grant prescriptive authority to qualified psychologists, and research has shown that these policies are associated with a reduction in suicides. In this study, we assess the cost-effectiveness of these policies in reducing suicide rates. This study used a Markov Model with a time horizon of 20 years to estimate the incremental net monetary benefit (INMB) of the policy from the societal perspective with a simulated cohort of 100,000 people. Transition probabilities and utilities were collected from the literature, and costs were assessed using a mixed macro-micro costing approach. Using this approach, we found that the 20-year INMB for the policy was estimated to be $12.81 million ($USD) per quality-adjusted life year (QALY). The probability of cost-effectiveness was greater than 50% at a willingness-to-pay threshold as low as $10,000 per QALY. The probability of cost-effectiveness was only modestly associated with the implementation costs of the policy, but was sensitive to the estimated effect of the policy intervention. The models estimated in this study support prescriptive authority for psychologists as a cost-effective strategy for reducing state-level suicide rates. A considerable amount of research is needed to understand the impact of this policy with finer granularity.

Charged Particle Pseudorapidity Distributions Measured with the STAR EPD

In 2018, in preparation for the Beam Energy Scan II, the STAR detector was upgraded with the Event Plane Detector (EPD). The instrument enhanced STAR’s capabilities in centrality determination for fluctuation measurements, event plane resolution for flow measurements, and in triggering overall. Due to its fine radial granularity, it can also be utilized to measure pseudorapidity distributions of the produced charged primary particles, in EPD’s pseudorapidity coverage of 2.15<|η|<5.09. As such a measurement cannot be done directly, the response of the detector to the primary particles has to be understood well. The detector response matrix was determined via Monte Carlo simulations, and corrected charged particle pseudorapidity distributions were obtained in Au + Au collisions at the center of mass collision energies sNN = 19.6 and 27.0 GeV using an iterative unfolding procedure. Several systematic checks of the method were also done.

Research on Personalized Recommendation Algorithm of Internet Platform Goods Based on Knowledge Graph

Personalized recommendation method is an effective means to filter out the information users need from a large amount of information, which is rich in practical value. Personalized recommendation methods are maturing, and many e-commerce platforms have been using different forms of recommendation methods with great success. In the recommendation systems of large-scale e-commerce platforms, traditional recommendation algorithms represented by collaborative filtering are modeled only based on users' rating data, and sparse user-project interaction data and cold start are two inevitable problems. The introduction of knowledge graphs in recommendation systems can effectively solve these problems because of their rich knowledge content and powerful relationship processing capability. In this paper, we study the personalized recommendation algorithm based on knowledge graph as auxiliary information, and use the temporal information of user-item interaction in the graph to model users' interests over time at a finer granularity, taking into account the problem of high training time cost of the model due to frequent updates of the knowledge graph when recommending to users dynamically. The article proposes the Interactive Knowledge-Aware Attention Network Algorithmic Model for Recommendations (IKANAM) and conducts comparison experiments on public datasets. The results show that the IKANAM recommendation algorithm can effectively improve the effectiveness of personalized recommendation of products on Internet platforms.

Low-loss microwave photonic switching for satellite communication.

For traditional switching architecture, packet switching performs fine granularity data packet forwarding, but its digital signal processing (DSP) has high power consumption (PC). All-optical switching provides rapid exchange of wavelength resources, which has coarse granularity. In scenarios where the PC is limited, such as broadband satcom, a switching architecture with lower PC and finer granularity than optical switching would be useful. In this paper, we propose a novel, to the best of our knowledge, low-loss microwave photonic switching architecture that can exchange subband signals across beams and frequency bands. The switching process is realized by exchanging optical carriers instead of payload signals, which does not degrade the signal power, guaranteeing the signal-to-noise ratio (SNR). We conducted a proof-of-concept experiment of 2 × 2 switching with two 1.2-GBaud quadrature phase-shift keying (QPSK) signals; an error vector magnitude (EVM) of or less than 13.87% is realized after forwarding. The proposed system has the advantages of low PC, high SNR, and fine granularity, and is very promising for flexible forwarding in future satcom systems.

Contrastive Learning Based Graph Convolution Network for Social Recommendation

Exploiting social networks is expected to enhance the performance of recommender systems when interaction information is sparse. Existing social recommendation models focus on modeling multi-graph structures and then aggregating the information from these multiple graphs to learn potential user preferences. However, these methods often employ complex models and redundant parameters to get a slight performance improvement. Contrastive learning has been widely researched as an effective paradigm in the area of recommendation. Most existing contrastive learning-based models usually focus on constructing multi-graph structures to perform graph augmentation for contrastive learning. However, the effect of graph augmentation on contrastive learning is inconclusive. In view of these challenges, in this work, we propose a contrastive learning based graph convolution network for social recommendation (CLSR), which integrates information from both the social graph and the interaction graph. First, we propose a fusion-simplified method to combine the social graph and the interaction graph. Technically, on the basis of exploring users’ interests by interaction graph, we further exploit social connections to alleviate data sparsity. By combining the user embeddings learned through two graphs in a certain proportion, we can obtain user representation at a finer granularity. Meanwhile, we introduce a contrastive learning framework for multi-graph network modeling, where we explore the feasibility of constructing positive and negative samples of contrastive learning by conducting data augmentation on embedding representations. Extensive experiments verify the superiority of CLSR’s contrastive learning framework and fusion-simplified method of integrating social relations.

Sample-Dependent Adaptive Temperature Scaling for Improved Calibration

It is now well known that neural networks can be wrong with high confidence in their predictions, leading to poor calibration. The most common post-hoc approach to compensate for this is to perform temperature scaling, which adjusts the confidences of the predictions on any input by scaling the logits by a fixed value. Whilst this approach typically improves the average calibration across the whole test dataset, this improvement typically reduces the individual confidences of the predictions irrespective of whether the classification of a given input is correct or incorrect. With this insight, we base our method on the observation that different samples contribute to the calibration error by varying amounts, with some needing to increase their confidence and others needing to decrease it. Therefore, for each input, we propose to predict a different temperature value, allowing us to adjust the mismatch between confidence and accuracy at a finer granularity. Our method is applied post-hoc, enabling it to be very fast with a negligible memory footprint and is applied to off-the-shelf pre-trained classifiers. We test our method on the ResNet50 and WideResNet28-10 architectures using the CIFAR10/100 and Tiny-ImageNet datasets, showing that producing per-data-point temperatures improves the expected calibration error across the whole test set.

Investigation of the Mechanical Properties of Iron Tailings Concrete Subjected to Dry-Wet Cycle and Negative Temperature.

This research investigates the effects of iron tailings content on the mechanical properties and durability of concrete under dry-wet cycling and negative temperature conditions (-10 °C), where iron tailings replace river sand at rates of 0%, 10%, 20%, and 30%. A variety of tests were conducted on the iron tailings concrete, including compressive strength, flexural strength, splitting tensile strength, mass loss, and relative dynamic modulus, and its pore characteristics were analyzed using low-field nuclear magnetic resonance (NMR) experiments. The results reveal that when 20% of the river sand was replaced with iron tailings, the concrete achieved optimal splitting strength, compressive strength, and flexural strength at 28 days, improving by 0.46 MPa, 3.14 MPa, and 0.41 MPa, respectively, compared to conventional concrete. Furthermore, the concrete containing this proportion of iron tailings demonstrated superior mechanical properties and durability in both negative temperature conditions and dry-wet cycling experiments. Due to the excellent physical and chemical properties of iron tailings, they enhance the performance of concrete when incorporated in appropriate quantities. The fine granularity of iron tailings helps to compensate for the granularity defects in concrete aggregates by filling internal voids, optimizing the pore structure, and improving the concrete's density and integrity. This enhances the concrete's mechanical properties and its resistance to external solutions and harmful ion penetration. Additionally, the active substances in iron tailings promote the hydration reaction of cement, leading to the formation of an increased amount of C-S-H gel and other hydration products in the cement system.

A review of video‐based rainfall measurement methods

AbstractAccurate and high spatiotemporal resolution rainfall observations are essential for hydrological forecasting and flood management, especially in urban hydrological applications. However, it is difficult for traditional rainfall gauges, weather radars, and satellites to accurately estimate rainfall while simultaneously capturing the spatial and temporal variability of rainfall well. In this context, video‐based rainfall measurement, a novel method, has the advantages of real‐time performance and low cost and may thus provide a new way to establish rainfall observation networks with high spatial and temporal resolution. In recent years, different algorithms have been developed to recognize raindrops and estimate rainfall from rainfall videos. It has been demonstrated that video‐based rainfall measurement methods can provide comprehensive rainfall information with fine spatial and temporal granularity. However, raindrop visibility and the depth of field effects are difficult to address. The motion blur effect of raindrops may result in substantial errors and uncertainties. A fundamental problem of video‐based rainfall measurements lies in locating raindrops and accurately calculating their actual size. Moreover, the effectiveness of deep learning‐based video rainfall measurement models is greatly influenced by the diversity of the training data. Therefore, enhancing the high robustness and accuracy of video‐based rainfall measurement algorithms and increasing the computational efficiency are paramount to further development, which are prerequisites for their application in practical rainfall monitoring and developing multicamera monitoring networks.This article is categorized under: Science of Water > Methods Science of Water > Hydrological Processes

Dynamic transformer ODEs for large-scale reservoir inflow forecasting

Forecasting incoming water demand is a critical step in efficient reservoir management and revenue optimization in large-scale cascade hydropower stations. It depends on multiple factors, such as weather conditions, grid dispatch, and electricity demand, and, in turn, facilitates a range of downstream decision-making, from natural hazards control and water ecology protection to power generation plans. Current efforts mainly rely on methodologies from statistical machine learning or deep neural networks to model the hydrological patterns from historical time series for inflow forecasting. However, existing models are restricted by short-term temporal dependencies and are prone to error accumulation issues due to the underlying autoregressive architecture. Meanwhile, most recent self-attention time-series models fail to achieve real-time inflow forecasting because of tremendous parameters and computational bottlenecks on learning long time series with fine granularity. We propose a novel framework, called DTODE (Dynamic Transformer Ordinary Differential Equations), for capturing nonlinear and non-stationary evolving patterns inherent in hydrological time series. Specifically, we present the dynamic self-attention mechanism combining transformer and ordinary differential equations that simultaneously captures long-range dependencies of observations from a dynamic system perspective. DTODE exploits a continuum of self-attention layers (instead of discrete counterparts) to learn the dynamics of multivariate time series while paying attention to the co-evolving time-related factors. Besides, our model is flexible in inferring the complex states at any time step, allowing us to forecast inflows at multiple time horizons. Comprehensive evaluations on real-world datasets show that DTODE significantly reduces forecasting errors compared to state-of-the-art inflow prediction systems.

Direct learning-based deep spiking neural networks: a review.

The spiking neural network (SNN), as a promising brain-inspired computational model with binary spike information transmission mechanism, rich spatially-temporal dynamics, and event-driven characteristics, has received extensive attention. However, its intricately discontinuous spike mechanism brings difficulty to the optimization of the deep SNN. Since the surrogate gradient method can greatly mitigate the optimization difficulty and shows great potential in directly training deep SNNs, a variety of direct learning-based deep SNN works have been proposed and achieved satisfying progress in recent years. In this paper, we present a comprehensive survey of these direct learning-based deep SNN works, mainly categorized into accuracy improvement methods, efficiency improvement methods, and temporal dynamics utilization methods. In addition, we also divide these categorizations into finer granularities further to better organize and introduce them. Finally, the challenges and trends that may be faced in future research are prospected.

Fast Progressive Differentiable Architecture Search based on adaptive task granularity reorganization

Shrinkage methods reduce the search space of a Differentiable Architecture Search (DARTS) by progressively discarding candidates, which accelerates the search speed. However, their shrinkage strategy suffers from the vulnerability of too fine task granularity. In other words, they drop only one of the least promising candidates per round of shrinkage, which is suboptimal in terms of performance and efficiency. In this study, we introduce the concept of Granular Computing (GrC) into the shrinkage method and present a Fast Progressive Differentiable Architecture Search (FP-DARTS) method. This method effectively reduces the computational complexity of each round of shrinkage, thereby improving the efficiency and performance of the algorithm. FP-DARTS can be divided into three stages: adaptive granularity division and selection, granular-channel performance evaluation, and progressive shrinkage. In the first stage, to reorganize the task granularity, we cluster the candidate operations into granular-channels and adaptively select the appropriate task granularity. We also propose a dynamic clustering strategy to avoid introducing additional computation. In the second stage, we train the architecture parameters to measure the potential of the granular-channels. In the third stage, to improve the stability of the shrinkage results, we introduce a channel annealing mechanism to smoothly discard unpromising granular-channels. We conducted systematic experiments on CIFAR-10 and ImageNet and achieved a test accuracy of 97.56% on CIFAR-10 with 0.04 GPU-days, and a test accuracy of 75.5% on ImageNet with 1.2 GPU-days. We also conducted experiments on the search space of NAS-Bench-201, and obtained test accuracies of 94.22, 73.07, and 46.23% for CIFAR-10, CIFAR-100 and ImageNet16-120, respectively. The above experimental results demonstrate that FP-DARTS achieves higher search speed and competitive performance compared to other state-of-the-art shrinkage methods and non-shrinkage methods.

Flexible transceiver for an access network: a multicarrier entropy loading approach

To cope with the extraordinary bandwidth demand, an optical access network based on point-to-multipoint (P2MP) topology has been widely applied. Recently, flexible transceivers have been investigated to maximize the optical power budget of optical network units for increasing the capacity of passive optical networks. In this work, we experimentally demonstrate a low-cost, high-bandwidth P2MP solution based on multicarrier modulations and probabilistic constellation shaping technology by reusing a 10G class directly modulated laser and avalanche photodetector for the first time to our knowledge. The proposed P2MP solution is capable of achieving fine granularity data rate adjustment from 12.5 Gb/s up to 63 Gb/s. We discuss the hardware implementation aspect of the proposed multicarrier entropy loading flexible transceiver. By means of a real-time transmission experiment, we show that the proposed P2MP solution can achieve a 29 dB power budget at a symmetric 53 Gb/s line rate, which makes it well positioned to support non-residential P2MP applications such as passive optical local area networks and fiber-to-the-office.