Abstract Data quality and budget are two major concerns in large-scale urban Mobile Crowdsensing (MCS) technologies. Traditional MCS research primarily measures data quality based on sensing coverage, without fully considering the importance of the data for downstream tasks. As a result, in Sparse Mobile Crowdsensing (SMCS), existing studies often focus on selecting the subregions most valuable for data inference, overlooking the ultimate goal of data collection, which is to support subsequent decision-making. With the rise of Embodied AI, a core challenge is how to actively collect key information that is most helpful for decision-making under conditions of sparse data and high costs. For example, in urban air quality monitoring, insufficient sensor deployment may cause critical polluted regions to go unobserved, undermining public health decisions. Similarly, sparse traffic data can lead to errors in autonomous driving systems, such as flawed route planning or hazard detection failures. As a feasible data collection paradigm for Embodied AI, existing SMCS subregion selection methods often ignore the relevance of data to tasks, thereby affecting decision-making in Embodied AI. To address this, we design a Sparse Mobile Crowdsensing active perception framework that selects the most valuable subregions for decision-making, considering the context of downstream tasks. Furthermore, existing SMCS subregion selection methods usually focus only on selecting the optimal subregion, whereas Embodied AI requires selecting a set of subregions. Choosing only the optimal subregion may overlook information overlap between subregions, leading to wasted data collection costs and reducing the perception efficiency of Embodied AI. This paper proposes an active acquisition strategy that takes into account parameter uncertainty and probabilistic margins, enabling the selection of an optimal set of subregions. Since this is an NP-hard problem, we approximate the proposed strategy using a greedy algorithm and provide performance guarantees for this approximation through theoretical proofs. Finally, we conduct experiments on three real-world datasets to validate the effectiveness of our proposed method. Experimental results show that when the data missing rate exceeds 60%, MBBALD starts to outperform other algorithms. On the traffic flow dataset, it achieves over 4% higher accuracy than the second-best algorithm at a 50% missing rate. Its advantage becomes more evident as the number of selected points increases, surpassing the second-best method by over 3% with more selected points.

Multi-interest recommendation constantly aspires to an oracle individual preference modeling approach, that satisfies the diverse and dynamic properties. Fueled by the deep learning technology, existing neural network (NN)-based recommender systems employ single-point or multi-point interest representation strategy to realize preference modeling,and boost the recommendation performance with a remarkable margin. However, as parameterized approximate functions, NN-based methods remain deficiencies with respect to the adaptability towards distinctive preference patterns cross different users and the calibration over the individual current intent. In this paper, we revisit multi-interest recommendation with the lens of stochastic process and Bayesian inference. Specifically, we propose to learn a distribution over functions to depict the individual diverse preferences rather than a unified function to approximate preference. Subsequently, the recommendation is encouraged with the uncertainty estimation which conforms to the dynamic shifting intent. Along these lines, we establish the connection between multi-interest recommendation and neural processes by proposing NP-Rec, which realizes the flexible multiple interests modeling and uncertainty estimation, simultaneously. Empirical study on 4 real world datasets demonstrates that our NP-Rec attains superior recommendation performances to several state-of-the-art baselines, where the average improvement achieves up to 13.94%.

The Internet of Things (IoT) enables the continuous collection of spatiotemporal data through mobile and edge devices. However, such data is often incomplete due to sensor failures or transmission issues, requiring imputation techniques to recover missing values. Effective imputation relies on accurate spatial and temporal information, which poses serious privacy risks when data is shared or processed externally. Existing privacy-preserving methods typically obfuscate spatial data and overlook temporal privacy, leading to degraded imputation accuracy. To address this, we propose a unified framework that integrates a novel spatiotemporal privacy mechanism directly into the imputation process. We propose a definition of spatiotemporal distance, which serves as the foundation for a novel spatiotemporal privacy protection mechanism. Building upon this, we integrate the privacy protection mechanism as a conditional component into the imputation algorithm, enabling accurate data imputation under privacy-preserving constraints. Experiments on six real-world datasets demonstrate superior performance in both imputation accuracy and privacy protection compared to baseline methods.

Traditional knowledge graphs (KGs) provide each entity with a unique embedding as a representation, which contains a lot of redundant information. Meanwhile, the space complexities of the KGs are positively related to the number of entities. In this work, we propose a hierarchical representation learning method, namely HRL, which is a parameter-efficient model where the number of model parameters is independent of dataset scales. Specifically, we propose a hierarchical model comprising a Meta Encoder and a Context Encoder to generate the representation of entities and relations. The Meta Encoder captures the common representations shared across entities, while the Context Encoder learns entity-specific representations. We further provide a theoretical analysis of model design by constructing a structural causal model (SCM) when completing a knowledge graph. The SCM outlines the relationships between nodes, where entity embeddings are conditioned on both common and entity-specific representations. Note that our model is designed to reduce model scale while maintaining competitive performance. We evaluate HRL on the knowledge graph completion task using three real-world datasets. The results demonstrate that HRL significantly outperforms existing parameter-efficient baselines, as well as traditional state-of-the-art baselines of similar scale.

Human preference alignment (HPA) aims to ensure Large Language Models (LLMs) responding appropriately to meet human moral and ethical requirements. Existing methods, such as RLHF and DPO, rely heavily on high-quality human annotation, which restrict the efficiency of iterative online model refinement. To address the inefficiencies of human annotation acquisition, iterated online strategy advocates the use of fine-tuned LLMs to self-generate preference data. However, this approach is prone to distribution bias, because of differences between human and model annotations, as well as modeling errors between simulators and real-world contexts. To mitigate the impact of distribution bias, we adopt the principles of adversarial training, framing a zero-sum two-player game with a protagonist agent and an adversarial agent. With the adversarial agent challenging the alignment of protagonist agent, we continuously refine the protagonist’s performance. By utilizing min-max equilibrium and Nash equilibrium strategies, we propose Indirect Online Preference Optimization (IOPO) mechanism that enables the protagonist agent to converge without bias while maintaining linear computational complexity. Extensive experiments across three real-world datasets demonstrate that IOPO outperforms state-of-the-art alignment methods in both offline and online scenarios, evidenced by standard alignment metrics and human evaluations. This innovation reduces the time required for model iterations from months to one week, alleviates distribution shifts, and significantly cuts annotation costs.

Existing graph neural network (GNN) methods are typically built upon the i.i.d. assumption, emphasizing the enhancement of the test performance for in-distribution (ID) data. However, there has been limited exploration of their adaptability to scenarios involving unknown distribution data. On the one hand, in real-world application scenarios, graph data often expands continuously with the acquisition of external knowledge, which means that new nodes with unknown categories may be added to the graph data. The gap between the new node distribution and the original node distribution can make existing GNN methods less effective. On the other hand, existing out-of-distribution (OOD) detection methods often rely on the softmax confidence score, which makes the OOD data suffer from overconfident posterior distributions. To address the above issues, we propose an Energy Propagation-based Graph Neural Network (EPGNN), which improves the OOD generalization ability by endowing GNN with the capacity to detect the OOD nodes in the graph. Specifically, we first construct GNN encoder to obtain node embedding that incorporates neighborhood structural information. Then, we design a plug-and-play energy-based OOD evaluator by assigning corresponding energy values to different nodes. Finally, we construct a plug-and-play structure-aware energy propagation module and joint alignment regularization, which make the node energy more flexible during the training process. Extensive experiments on benchmark datasets demonstrate the superiority of our method.

Urban fine-grained data map inference, leveraging information from coarse-grained maps, has emerged as a significant area of research due to the growing complexity and data heterogeneity in urban environments.Existing methods have a priori assumption that a coarse-grained data map, one fixed-size granularity, transforms into a fine-grained data map, also one fixed-size granularity.However, in actual scenarios, the collected coarse-grained data maps are often incomplete and have significantly distinct granularities in various urban areas, which results in incomplete heterogeneous data, i.e., multi-granularity data maps in terms of spatial information.Meanwhile, different granularity data maps are needed for various urban downstream tasks, which is a multi-task problem.To that end, this paper proposes a novel framework, a multi-granularity super-resolution data map inference framework (MGSR), designed to harness spatio-temporal information to transform incomplete coarse-grained multi-granularity data maps into fine-grained multigranularity data maps.Specifically, we design a granularity alignment network to align multi-granularity information and address missing data on each granularity data map by leveraging the other granularity data maps with a well-designed self-supervised task.Then, we introduce a feature extraction network to capture spatiotemporal dependencies and extract features.Finally, we devise a recurrent super-resolution network with shared parameters to infer multi-granularity data maps.We conduct extensive experiments on three real-world benchmark datasets and demonstrate that MGSR significantly outperforms the state-of-the-art methods for multigranularity urban data map inference and reduces RMSE and MAE by up to 40.1% and 50.3%, respectively.

Spatial crowdsourcing is drawing much attention with the rapid development of mobile Internet. Achieving efficient crowdsourcing task assignment involves not only maximizing the earnings of workers but also balancing the preferences of users or customers. Users often express preferences for specific workers or conditions, such as particular drivers, delivery personnel, or service providers. To address this challenge, we investigate the future-aware balanced preference (FABP) problem. This problem aims to maximize the profits of global workers while simultaneously considering the preferences of both parties to ensure bilateral satisfaction. To address the FABP problem, we propose the learning to match (LTM) algorithm. This algorithm utilizes online reinforcement learning that considers both immediate profits and long-term rewards. It acknowledges the significance of task assignment decisions in relation to the spatial distribution of future drivers, which in turn affects subsequent decisions. The LTM algorithm generates future-aware preference lists using learned driver state values and guides the subsequent matching. Additionally, we present the real-time preference-based matching (RTPM) algorithm, which is a real-time matching algorithm that enables substitutions based on preference lists when a more preferred matching pair becomes available. This enhances the efficiency and fairness of real-time task assignment in dynamic environments, while simultaneously meeting the needs of passengers and drivers. Our extensive experiments on both real and synthetic datasets validate the effectiveness of our proposed algorithms, demonstrating a noteworthy improvement of up to 11.8% and an average increase of 4.7% compared to benchmark algorithms.

Most dataset distillation methods struggle to accommodate large-scale datasets due to their substantial computational and memory requirements. Recent research has begun to explore scalable disentanglement methods. However, there are still performance bottlenecks and room for optimization in this direction. In this paper, we present a curriculum-based dataset distillation framework aiming to harmonize performance and scalability. This framework strategically distills synthetic images, adhering to a curriculum that transitions from simple to complex. By incorporating curriculum evaluation, we address the issue of previous methods generating images that tend to be homogeneous and simplistic, doing so at a manageable computational cost. Furthermore, we introduce adversarial optimization towards synthetic images to further improve their representativeness and safeguard against their overfitting to the neural network involved in distilling. This enhances the generalization capability of the distilled images across various neural network architectures and also increases their robustness to noise. Extensive experiments demonstrate that our framework sets new benchmarks in large-scale dataset distillation, achieving substantial improvements of 11.1% on Tiny-ImageNet, 9.0% on ImageNet-1K, and 7.3% on ImageNet-21K. Our distilled datasets and code are available at https://github.com/MIV-XJTU/CUDD.

Mobile crowdsensing (MCS) is a paradigm for data collection with the limitation of budgets and worker availability. The central strategy of MCS is recruiting workers to sense a part of data and subsequently infer the unsensed data. To infer unsensed data, prior research has proposed several algorithms that do not require historical data, but their inference accuracy is very limited. More effective works are training a model with sufficient historical data. However, such methods can’t infer data with few to none historical data. A more promising strategy is training models from other similar datasets that have been sensed. However, such datasets are different in terms of sensing locations, numbers of sensed data and data types. Such variance introduces the complex issue of integrating knowledge from these datasets and then training inference models. To solve these, we propose a meta-knowledge based multiple data inference framework named META-MCS. In META-MCS, we propose a similarity evaluation model TMFS. Following this, we cluster similar datasets and train generalized models for each cluster. Finally, META-MCS selects an appropriate model to infer unsensed data. We validate our proposed methods through extensive experiments using ten different datasets, which substantiate the effectiveness of our framework.

In this work, we present TriMLP as a foundational MLP-like architecture for the sequential recommendation, simultaneously achieving computational efficiency and promising performance. First, we empirically study the incompatibility between existing purely MLP-based models and sequential recommendation, that the inherent fully-connective structure endows historical user–item interactions (referred as tokens) with unrestricted communications and overlooks the essential chronological order in sequences. Then, we propose the MLP-based Triangular Mixer to establish ordered contact among tokens and excavate the primary sequential modeling capability under the standard auto-regressive training fashion. It contains (1) a global mixing layer that drops the lower-triangle neurons in MLP to block the anti-chronological connections from future tokens and (2) a local mixing layer that further disables specific upper-triangle neurons to split the sequence as multiple independent sessions. The mixer serially alternates these two layers to support fine-grained preferences modeling, where the global one focuses on the long-range dependency in the whole sequence, and the local one calls for the short-term patterns in sessions. Experimental results on 12 datasets of different scales from 4 benchmarks elucidate that TriMLP consistently attains favorable accuracy/efficiency tradeoff over all validated datasets, where the average performance boost against several state-of-the-art baselines achieves up to 14.88%, and the maximum reduction of inference time reaches 23.73%. The intriguing properties render TriMLP a strong contender to the well-established RNN-, CNN-, and Transformer-based sequential recommenders. Code is available at https://github.com/jiangyiheng1/TriMLP .

This paper proposes an improved YOLOv5 deep learning algorithm to efficiently detect foreign objects in wireless power transmission (WPT) systems based on infrared images, in response to the presence or dropping of metallic foreign objects in wireless charging systems. First, the model is made lighter and the detection speed is improved by replacing the backbone network CSPDarkNet53 with MobileNetv3. At the same time, the ECA attention mechanism is added to make the network pay more attention to the parameter relationship information and localization information of the model, improving the feature extraction capability of the network. Secondly, the original loss function CIoU of the network is replaced with WIoU, which makes the model converge better and improves the precision of the algorithm detection. Finally, the improved YOLOv5 model was validated on the collected infrared image datasets for the detection of metallic foreign objects, with an increase in precision from 96.2% to 98.3%, recall from 96.5% to 98.8%, overall mAP to 99.5%, and the detection speed improved by nearly 15%, successfully enabling the detection of foreign objects in wireless power transmission systems via infrared images.

In the paper a position constraint algorithm is proposed to achieve high power density, based on the thermal resistance model of a half-bridge LLC system under natural convection. An LLC system with resonant frequency of 1MHz, input voltage of 290-330V, output voltage of 48V and power of 1kW is designed by using the third generation of power semiconductor device GaN. The electrical characteristics of the circuit are verified by simulation software. The loss of the system and the heat dissipation path of the LLC system with planar winding are analyzed, and the thermal resistance model of the system is established. Based on the thermal resistance model of the system, the heat dissipation power of each device is reduced, and the simplified thermal resistance model of each device is established. The device position constraint algorithm is established in a complex system without integration and differentiation. The simulation results show that in the LLC system under the position constraint algorithm, each device operates within the required temperature range, the error between the simulated value and the calculated value is within 5%, and the power density reaches 8.23W/cm3.

This paper presents a comprehensive analysis from a theoretical perspective, showcasing the development of a Multiphysics simulation that incorporates both electrical and magnetic components to investigate the energy conversion of a bidirectional DAB converter. Various phase shift modulation control methods are proposed and thoroughly validated through simulations. Notably, the extended phase shift modulation scheme successfully minimizes the inductor peak current and reactive power of the DAB converter. Beyond electrical circuit simulations, the study delves into the design and simulation of an 800V high-frequency transformer using COMSOL for the DAB circuit. In-depth testing, including open circuit and short circuit tests, is conducted to derive the equivalent circuit. Switching and conduction losses are also estimated based on datasheet specifications. Ultimately, the research culminates in the simulation and comparative analysis of converter efficiency across different modulation strategies. The findings shed light on the most effective approaches for optimizing the DAB converter's performance and energy conversion capabilities.

The success of deep learning relies heavily on large and diverse datasets, but the storage, preprocessing, and training of such data present significant challenges. To address these challenges, dataset distillation techniques have been proposed to obtain smaller synthetic datasets that capture the essential information of the originals. In this paper, we introduce a Sparse Parameterization for Epitomic datasEt Distillation (SPEED) framework, which leverages the concept of dictionary learning and sparse coding to distill epitomes that represent pivotal information of the dataset. SPEED prioritizes proper parameterization of the synthetic dataset and introduces techniques to capture spatial redundancy within and between synthetic images. We propose Spatial-Agnostic Epitomic Tokens (SAETs) and Sparse Coding Matrices (SCMs) to efficiently represent and select significant features. Additionally, we build a Feature-Recurrent Network (FReeNet) to generate hierarchical features with high compression and storage efficiency. Experimental results demonstrate the superiority of SPEED in handling high-resolution datasets, achieving state-of-the-art performance on multiple benchmarks and downstream applications. Our framework is compatible with a variety of dataset matching approaches, generally enhancing their performance. This work highlights the importance of proper parameterization in epitomic dataset distillation and opens avenues for efficient representation learning. Source code is available at https://github.com/MIV-XJTU/SPEED.

Readers of novels need to identify and learn about the characters as they develop an understanding of the plot. The paper presents an end-to-end automated pipeline for literary character identification and ongoing work for extracting and comparing character representations for full-length English novels. The character identification pipeline involves a named entity recognition (NER) module with F1 score of 0.85, a coreference resolution module with F1 score of 0.76, and a disambiguation module using both heuristic and algorithmic approaches. Ongoing work compares event extraction as well as speech extraction pipelines for literary characters representations with case studies. The paper is the first to my knowledge that combines a modular pipeline for automated character identification, representation extraction and comparisons for full-length English novels.

An omnidirectional wireless power transfer (WPT) system using three-phase frustum-shaped transmitting coil is designed. In the system, the receiving coil is able to receive the power at any position and any angle within the charging space. The three-phase frustum-shaped transmitting coil is consisted of three half frustum-shaped coils overlapped 60°in turn. The system is powered-up by a three-phase power source, which generates a uniformly rotating magnetic field in the space to achieve omnidirectional wireless charging with high degree of freedom. Based on the circuit principle, the equivalent circuit model of the system is established, the output power is calculated, and the main factors affecting the charging efficiency are analyzed. Then, using Maxwell, a three-phase positive sequence control strategy is designed through finite element simulation, and uniform-intensity magnetic field can be obtained in the space. Mutual inductance is used to indicate the influence of the receiving coil position on the output characteristics of the system. Finally, a three-phase WPT system prototype is built, and the transmission performance of the system is tested when the receiving coil rotates and moves longitudinally inside the frustum-shaped transmitting coils. The experimental results show that the maximum charging power is 12W and the maximum charging efficiency is up to 55.2%, and the fluctuation of the charging efficiency is 9.5% as the receiving coil moving and rotating within the charging space.

The prosperity of industrialization urges modern ride-sharing platforms to gain profit from efficient management of their resources. Although ride-sharing allows sharing costs and promotes the traffic efficiency by making better use of vehicle capacities, dealing with large amounts of online taxi orders is an inevitable challenge in the current transportation systems, where all drivers have to make immediate and irrevocable decisions about whether to accept current order in a parallel way. Furthermore, in order to achieve global fairness, it is critical for an algorithm to function whenever the first order gets on-line without any observation stage. In this paper, we formulate this online user selection problem as a prophet inequality for independent identically distributed random variables from an unknown distribution. We construct a sample set to avoid the observation stage in an online decision process. Considering the driver-centered ride-sharing scenario, a route schedule algorithm and a sample-driven algorithm with a guarantee of lower bound are proposed to concurrently guide taxi drivers to accept taxi orders and achieve global fairness at the meantime. Finally, we conduct extensive evaluations based on three real-world data sets. The results verify the effectiveness of our proposed algorithm on improving the overall profit, increasing accepted orders and reducing the unoccupied time of the vehicle under the valid ride-sharing constraints.

For metal foreign object detection (MFOD) in wireless power transfer (WPT) system, the electromagnetic induction coil detection method has the advantages of simple structure, strong environmental adaptability, and low cost. Whether it is a passive detection scheme or an active detection scheme, the detection effect is closely related to the mutual inductance between the detection coil and the metal foreign object. Thus, the structure of the detection coil needs to be optimized. It is very meaningful to optimize the structure of the detection coil for the purpose of improving mutual inductance. In this paper, a rack-shaped detection coil is designed. Firstly, theoretical analysis is used to compare the mutual inductance between the single-turn rack and the circular metal foreign object and the mutual inductance between the rectangular detection coil and the circular metal foreign object. Secondly, simulation analysis is used to compare the mutual inductance between the rack coil pair and the metal foreign object and the mutual inductance between the rectangular coil pair and the metal foreign object. Finally, experiment shows that the single-layer rack coil can effectively detect 1 RMB coin and double-layer rack coil can accurately locate the coin on the basis of effective detection when the WPT system power is 1.5kW and the vertical distance between the detection coil and the primary coil is 15mm. 1 jiao RMB coin is the minimum size of the object that can be detected accurately by the detection system.