Short-term traffic speed prediction under limited historical data is crucial for intelligent transportation systems, enabling real-time traffic management and congestion mitigation. However, existing methods relying on long lookback window to forecast struggle to capture scenarios where the current traffic state depends more on recent patterns than on distant historical data. To address this challenge, we propose the Spatiotemporal Visibility Graph Network (STVGN), a novel framework that combines visibility graph theory with Graph Neural Networks (GNN). The core of STVGN is the STVGN-Embedding layer, which is designed to enhance short-term spatiotemporal feature representation. This layer leverages the visibility graph’s ability to capture transient patterns and integrates GNNs to model intrinsic relationships within the visibility graph derived from time series data. In Combination with a transformer architecture, STVGN-Embedding extracts complex spatiotemporal features, while the transformer uncovers inherent relationships between temporal and spatial dimensions. To the best of our knowledge, this is the first study to introduce visibility graph theory as an embedding layer within a transformer framework. Experiments on three benchmark datasets, covering urban and freeway traffic scenarios, demonstrate STVGN’s effectiveness, achieving state-of-the-art performance with improvements of 4.8%–16.8% in RMSE and 8.4%–13.3% in MAE over existing SOTA methods. These results highlight the potential of visibility graph-based embeddings to address challenges posed by limited historical data and to capture intricate traffic patterns.

Recent advancements have demonstrated the effectiveness of transformer models in framing decision-making as a sequence modeling task, especially noted for their scalability and efficiency within offline reinforcement learning contexts. Nonetheless, the extension of these models to the realm of Multi-Agent Reinforcement Learning (MARL) presents formidable obstacles, including credit assignment among individual agents while ensuring decentralized policy execution for practical deployment. This paper introduces Counterfactual Multi-agent Decison Transformer (CFMADT), a novel transformer-based approach for offline MARL within the Centralized Training with Decentralized Execution (CTDE) paradigm. CFMADT employs separate transformers for global and local trajectory modeling, using counterfactual reasoning to estimate individual agent contributions and facilitate effective credit assignment. The evaluation results on several complex offline MARL benchmarks, e.g., StarCraft Multi-Agent Challenge (SMAC) , SMAC v2 and Flatland, showcase that our method achieves exceptional performance.

Recent advancements in diffusion policies have demonstrated promising performance in decision-making tasks. To align these policies with human preferences, a common approach is incorporating Preference-based Reinforcement Learning (PbRL) into policy tuning. However, since preference data is practically collected from populations with different backgrounds, a key challenge lies in handling the inherent uncertainties in people's preferences during policy updates. To address this challenge, we propose the Diff-UAPA algorithm, designed for uncertainty-aware preference alignment in diffusion policies. Specifically, Diff-UAPA introduces a novel iterative preference alignment framework in which the diffusion policy adapts incrementally to preferences from different user groups. To accommodate this online learning paradigm, Diff-UAPA employs a maximum posterior objective, which aligns the diffusion policy with regret-based preferences under the guidance of an informative Beta prior. This approach enables direct optimization of the diffusion policy without specifying any reward functions, while effectively mitigating the influence of inconsistent preferences across different user groups. We conduct extensive experiments across both simulated and real-world robotics tasks, and diverse human preference configurations, demonstrating the robustness and reliability of Diff-UAPA in achieving effective preference alignment. The project page is available at https://github.com/mr20010112/Diff_UAPA.

In dense retrieval, embedding long texts into dense vectors can result in information loss, leading to inaccurate query-text matching. Additionally, low-quality texts with excessive noise or sparse key information are unlikely to align well with relevant queries. Recent studies mainly focus on improving the sentence embedding model or retrieval process. In this work, we introduce a novel text augmentation framework for dense retrieval. This framework transforms raw documents into information-dense text formats, which supplement the original texts to effectively address the aforementioned issues without modifying embedding or retrieval methodologies. Two text representations are generated via large language models (LLMs) zero-shot prompting: question-answer pairs and element-driven events. We term this approach QAEA-DR: unifying question-answer generation and event extraction in a text augmentation framework for dense retrieval. To further enhance the quality of generated texts, a scoring-based evaluation and regeneration mechanism is introduced in LLM prompting. Our QAEA-DR model has a positive impact on dense retrieval, supported by both theoretical analysis and empirical experiments.

The emergence of large language models (LLMs) has revolutionized the way we interact with graphs, leading to a new paradigm called GraphLLM. Despite the rapid development of GraphLLM methods in recent years, the progress and understanding of this field remain unclear due to the lack of a benchmark with consistent experimental protocols. To bridge this gap, we introduce GLBench, the first comprehensive benchmark for evaluating GraphLLM methods in both supervised and zero-shot scenarios. GLBench provides a fair and thorough evaluation of different categories of GraphLLM methods, along with traditional baselines such as graph neural networks. Through extensive experiments on a collection of real-world datasets with consistent data processing and splitting strategies, we have uncovered several key findings. Firstly, GraphLLM methods outperform traditional baselines in supervised settings, with LLM-as-enhancers showing the most robust performance. However, using LLMs as predictors is less effective and often leads to uncontrollable output issues. We also notice that no clear scaling laws exist for current GraphLLM methods. In addition, both structures and semantics are crucial for effective zero-shot transfer, and our proposed simple baseline can even outperform several models tailored for zero-shot scenarios. The data and code of the benchmark can be found at https://github.com/NineAbyss/GLBench.

Offline reinforcement learning (RL) enables learning policies from precollected datasets without online data collection. Although it offers the possibility to surpass the performance of the datasets, most existing offline RL algorithms struggle to compete with behavior cloning policies in many dataset settings due to trading off policy improvement and additional regularization to address the distributional shift issue. In many cases, if one can imitate a sequence of suboptimal subtrajectories in data and properly“stitch”them toward reaching an ideal future state, it may potentially result in a more reliable policy while avoiding difficulties that present in typical value-based offline RL algorithms. We borrow the idea of curriculum learning to embody the above intuition. We construct a curriculum that progressively imitates a sequence of suboptimal trajectories conditioned on a series of carefully constructed future states and cumulative rewards as goals. The suboptimal trajectories gradually guide policy learning toward reaching the ideal goal states. We name our algorithm curriculum goal-conditioned imitation (CGI). Experimental results show that CGI achieves competitive performance against state-of-the-art offline RL algorithms, especially for challenging tasks with long horizons and sparse rewards.

This paper proposes an educational game design framework named the Knowledge-based Personality, Emotion, and Action Cues (KPEAC). It aims to naturally enhance the user's intrinsic motivation and learning effectiveness by employing these novel Anthropomorphic Design Cues (ADC). As an example of this framework, we developed Chemical Life, a chemistry education game, where existing atoms and molecules act as characters with assigned personality traits, emotions, and reactions according to chemical rules. For instance, sodium atoms with unstable properties exhibit neuroticism, while oxygen molecules, which readily engage in various reactions, demonstrate openness. These characters engage in dialogues generated with the assistance of a large language model (LLM). Players initially interact with these chemical characters, immersing themselves in their personalities and emotions, before guiding them to perform the appropriate chemical reactions. Preliminary experimental results indicate that this framework stimulates users’ intrinsic motivation and facilitates knowledge acquisition in a proactively goal-driven manner. It also provides new directions for the application of ADC and LLMs in the field of educational games.

Parallel Dance is a social game on campus public screens based on motion capture, which provides a platform for students to find new friends. We propose a multiplayer game model, in which players participate in the same game cross space and time. We focus on helping students overcome social shame, lack of self-confidence, and language barriers in this social game. We designed and implemented Parallel Dance, and deployed the game on campus public screens which are networked by cloud server. Participants can gain not only fun in the game, but also friendship.

Although live streaming has grown increasingly popular, the interactive channels between the streamer and their viewers stays limited - mostly via live chat and virtual gifting. To improve this situation, thanks to the rise of Virtual Youtuber (Vtuber), we design a new game model - Multiviewer and Vtuber Extreme Asymmetric Game (MVEAG) - that profoundly enhances the interactivity of virtual live streaming. In the MVEAG model, massive viewers interact with the Vtuber streamer in the very game scene where they play together, but with extremely asymmetric roles. The Vtuber plays the main role whereas the viewers play supplementary roles. Both roles are indispensable in achieving the game objective so that the Vtuber has to establish different interactive strategies with their viewers in the gameplay. We design the Gulliver’s Game as an example of MVEAG model, in which we demonstrate spontaneous and complex in-game interactive behaviors, when multiviewers experience direct collaboration and confrontation with the streamer.

An increasing amount of location-based service data is being accumulated and helps to study urban dynamics and human mobility. Location embedding generated from human mobility trajectories has become a popular topic to understand urban functionality, and could be applied as essential resources to various downstream tasks. Existing location embedding methods are mostly tailored for specific problems that are taken place within a small group of areas. Downscaling the spatial resolution makes existing approaches suffer from extensive computational cost and significant data sparsity. We propose to learn finegrained location representations through a GCN-aided skip-gram model named GCN-L2V by considering both spatial adjacency and human mobility. With a flow graph and a spatial graph, it embeds context information into vector representations. GCN-L2V is able to capture relationships among locations and provides a better notion of similarity in spatial environment. Quantitative experiments and case studies empirically demonstrate that representations learned by GCN-L2V are effective.

In optical imaging systems, the aberration is an important factor that impedes realising diffraction‐limited imaging. Accurate wavefront sensing and control play important role in modern high‐resolution optical imaging systems nowadays. In this study, a simple model‐based phase retrieval algorithm is proposed for accurate efficient wavefront sensing with high dynamic range. In the authors’ algorithm, a wavefront is represented by the Zernike polynomials, and the Zernike coefficients are solved by the least‐squares‐based non‐linear optimisation method, i.e. the Lederberg–Marquardt algorithm, with multiple phase‐diversity images. The numerical results show that the proposed algorithm is capable of retrieving wavefront with a large dynamic range up to seven wavelength and robust to noise. In comparison, the proposed algorithm is more efficient than the existing model‐based technique and more accurate than existing Fourier ‐ transformation‐based iterative techniques.

MOTIVATION: The concept of synergy between two agents, over a century old, is important to the fields of biology, chemistry, pharmacology and medicine. A key step in drug combination analysis is the selection of an additivity model to identify combination effects including synergy, additivity and antagonism. Existing methods for identifying and interpreting those combination effects have limitations. RESULTS: We present here a computational framework, termed response envelope analysis (REA), that makes use of 3D response surfaces formed by generalized Loewe Additivity and Bliss Independence models of interaction to evaluate drug combination effects. Because the two models imply two extreme limits of drug interaction (mutually exclusive and mutually non-exclusive), a response envelope defined by them provides a quantitatively stringent additivity model for identifying combination effects without knowing the inhibition mechanism. As a demonstration, we apply REA to representative published data from large screens of anticancer and antibiotic combinations. We show that REA is more accurate than existing methods and provides more consistent results in the context of cross-experiment evaluation. AVAILABILITY AND IMPLEMENTATION: The open-source software package associated with REA is available at: https://github.com/4dsoftware/rea. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

The hospital system is a complex service system in which patients and hospitals interact and make their decisions based on bounded rationality and information. In this paper, we develop a generative agent-based model to simulate the behavior of a hospital service system. Our model combines agent-based simulation and queueing models to mimic the hospital service processes. Our goal is to simulate and understand the growth and size distribution of hospitals. This simulation model includes agents for supply elements (i.e., hospitals with different resources and expansion strategies) and demand elements (i.e., patients with different preferences for their hospital selections) in a hospital service system. Three important questions are investigated: 1) what is the emergent size distribution of hospitals? 2) what key factors influence the size distribution? and 3) how sensitive is the size distribution to these key factors? Simulation results show that the size distribution is neither power law nor lognormal. Rather, the distribution is leptokurtic, and more skewed than normal but less skewed than lognormal. This result contradicts those in extant literature on the size distributions of human and natural systems, including cities, firms, power-grids, and citation network. We conduct a set of experiments to identify the generative mechanisms for the hospital size distribution and to test the robustness of the results. The model was validated empirically by using a U.S. hospital size dataset.

This paper introduces and solves a decision-making problem under the context of consensus control with failure. We study an optimal consensus control problem in which n autonomous agents try to arrive at a target at the same time. One of the agents suddenly fails and the rest n - 1 agents can either wait or abandon the failed agent. If they wait, they must slow down and delay the consensus time. If they abandon the failed agent, they can reach consensus earlier at the cost of losing one agent at consensus. This cost is an added delay to the consensus time. The decision problem is to decide whether to wait or abandon and, if abandon, when? To solve this problem, we derive analytical expressions and establish structural properties for target distance functions. We use numerical examples and simulation examples to demonstrate the applications of the derived formulas and results.

Wireless medium access control (MAC) protocols usually provide reliability in the presence of packet errors. The efficiency of these reliability mechanisms is generally low since they require the entire packet to be retransmitted even though only parts of it may have been corrupted. To address this issue, this paper presents an efficient mechanism for the recovery of packets corrupted due to both channel errors and collisions. The proposed mechanism first determines the cause of the errors. Next, the symbols with errors are isolated by using the error vector magnitude (EVM) of received symbols as the feature for detection. Using explicit feedback about which blocks of symbols have errors, only the erroneous blocks are then retransmitted. Our results show that the proposed mechanism increases the efficiency of the MAC protocol by providing higher throughput.

There are two causes of packet losses during a wireless transmission: losses caused by collisions and losses caused by poor channel conditions. The throughput and spatial reuse of IEEE 802.11 based wireless networks, as well as the effectiveness of the rate adaptation algorithms they use, is adversely affected by their inability to determine the real cause of a packet loss. To address this issue, this paper proposes a mechanism based on Error Vector Magnitude (EVM) to discern random channel errors from collisions in wireless networks. The proposed mechanism is based on first developing an analytic model to characterize the EVM of a packet in the presence and absence of a collision. A threshold based classifier is then proposed that selects the threshold value such that the crossover error rate is achieved. Simulation results are presented to demonstrate the accuracy of the proposed collision detection mechanism.

Service is about value cocreation between customers and providers. Cocreation builds on human networking: people connecting fluidly with each other as customers, providers, and resources to pursue common values. This paper develops a new analysis of service value networks, building on a previously presented hypernetwork model to study how people can scale their value cocreation up to span the entire population (domain), down to meet individual needs, and transformationally to breed new business designs. The new analysis reflects the convergence of social networks and e-commerce, and the evolution of physical products toward incorporating services to users into them (such as the apps and digital resources on the iPod, iPhone, and iPad). The hypernetwork model analyzes human networks that overlay multidimensionally, such as the Internet community itself. These properties extend the previous research results on random graphs and semiregular networks. A simulation study helps verify the hypernetworking analysis.

In this paper, we analyze optimal scheduling of a tree-like multicluster tool with single-blade robots and constant robot moving times. We present a recursive minimal cycle time algorithm to reveal a multi-unit resource cycle for multicluster tools under a given robot schedule. For a serial-cluster tool, we provide a closed-form formulation for the minimal cycle time. The formulation explicitly provides the interaction relationship among clusters. We further present decomposition conditions under which the optimal scheduling of multicluster becomes much easier and straightforward. Optimality conditions for the widely used robot pull schedule are also provided. An example from industry production is used to illustrate the analytical results. The decomposition and optimality conditions for the robot pull schedule are also illustrated by Monte Carlo simulation for the industrial example.

In semiconductor manufacturing, finding an efficient way for scheduling a multicluster tool is critical for productivity improvement and cost reduction. This two-part paper analyzes optimal scheduling of multicluster tools equipped with single-blade robots and constant robot moving times. In this first part of the paper, a resource-based method is proposed to analytically derive closed-form expressions for the minimal cycle time of two-cluster tools. We prove that the optimal robot scheduling of two-cluster tools can be solved in polynomial time. We also provide an algorithm to find the optimal schedule. Examples are presented to illustrate the proposed approaches and formulations.