Distance-decay models are fundamental to accessibility modeling; yet their alignment with actual travel behavior remains insufficiently examined in empirical terms. To help address this gap, we propose a dual-component Tanner–Gaussian decay model that seeks to mitigate two key limitations of traditional accessibility frameworks by simultaneously describing the phenomena of full decay and local peaks. The model is calibrated using survey data from an urban park in Hangzhou, China, and subsequently assessed on two additional datasets from Wuhan and Shanghai. Results indicate that: (1) traditional Gaussian functions may overestimate short-distance and underestimate long-distance accessibility, while Tanner functions tend to capture long-tail decay more effectively and appear more suitable for long-distance accessibility estimation; (2) the proposed dual-component model performs favorably in large samples, though its stability appears sensitive to sample size. By comparing the application of multiple accessibility models across three datasets, we highlight how their suitability varies depending on the specific context. This comparison may offer a reference that could assist designers in identifying underserved areas and supporting more equitable access to urban green spaces in the context of rapid urbanization.

We study a data-driven multi-item newsvendor problem with resource constraints. Demand of each item depends on exogenous features and a random shock. The objective is to obtain a data-driven ordering decision that minimizes the inventory cost. We assume that the firm has no prior knowledge on demand distributions, but has access to past demand samples and related feature information. We adopt local learning methods that approximate the objective function using weight sample average in which the weights can be computed by the k-nearest neighbors (kNN) and kernel regression methods. Such an approximation enables us to determine the order quantities directly from historical data. We then analytically derive the effectiveness of our proposed methods by showing their asymptotic optimality. That is, the ordering decisions derived from our methods and the corresponding costs converge to the optimal ones that are obtained by assuming known demand distributions. We also provide explicit performance bounds for our proposed methods in terms of the number of items, the feature dimensions and the sample size. Finally, numerical studies show several observations: (1) our methods outperform three widely-used baselines: Sample Average Approximation (SAA), Empirical Risk Minimization (ERM), and Predict-then-Optimize (PTO), showing consistent advantages across item numbers, service levels, and constraint tightness, as long as the feature dimension is not extremely large; (2) the cost gaps of all methods are decreasing as the resource constraint becomes tighter and these methods perform closely when the resource is very limited; (3) the Empirical Risk Minimization (ERM) method fails to converge and is far away from optimality, even when the sample is relatively large. In a real case study of one of the biggest e-commerce companies in China, our proposed methods outperform all the others, and the inventory costs are significantly reduced.

Product returns are prevalent in practice. Many retailers provide lenient free return policies but with specific return window within which customers are allowed to return products. Motivated by this phenomenon, we consider a single-product online learning and pricing problem with stochastic product returns. A salient feature is that the demand function, depending on price and return window decisions, is initially unknown and must be learned on the fly. The retailer thus faces the classic exploration-exploitation trade-off. Moreover, we consider an inventory constraint, introducing an additional trade-off between earning revenue and managing inventory. We propose a modeling framework to integrate pricing and return window decisions, and develop a deterministic fluid model that serves as the full-information benchmark. To tackle the learning problem, we design a novel nonparametric learning algorithm that seamlessly integrates inverse stochastic gradient descent (SGD) and Upper Confidence Bound (UCB) methods. Under mild assumptions on demand and revenue functions, we establish a regret upper bound for our learning algorithm as O ( W T log T ) , where W denotes the number of return window candidates and T denotes the time horizon. This result aligns with lower bounds established in both online pricing and multi-armed bandit (MAB) literature. Numerical experiments are conducted to verify the effectiveness and robustness of our algorithm across various environments. From an operational standpoint, retailers can use our learning framework as a decision-support tool to identify the optimal price and return window.

Nearly zero-energy buildings are increasingly evolving from conventional energy consumers into systems with both demand and on-site supply functions. However, the operational changes associated with shifts in supply-demand dominance remain insufficiently understood. This study investigated a nearly zero-energy office building. Under unified boundary conditions, comparative models of reference and designed buildings were established to reconstruct hourly building energy demand and on-site renewable energy supply profiles. The transition of supply-demand relationships was evaluated from four aspects: structural characteristics, supply-demand status, temporal coupling, and techno-economic performance. Results show that the annual energy demand was reduced by 40.8%, while the supply-demand ratio increased from 0 to 1.44, indicating a shift from external dependence to annual net surplus. Meanwhile, the end-use load structure was reshaped, with heating and air-conditioning loads significantly decreasing and internally driven loads relatively increasing, transforming the system from a single-dominant structure to a multi-load interactive structure. Further analysis reveals that although annual energy surplus was achieved, only 54.9% of the on-site supply was effectively utilized due to temporal mismatch between supply profiles and demand patterns, with a mismatch index of 0.352. In addition, under the present annual-scale engineering boundary, the unit energy investment intensity of demand-side pathways was lower than that of supply-side pathways by a factor of 1.82. These findings indicate that optimization priorities for buildings with high renewable penetration should move beyond demand reduction toward coordinated improvement of structural balance, temporal matching, and pathway configuration. This study provides a new perspective for the design and operation of nearly zero-energy and zero-carbon building energy systems.

Despite strong results on many tasks, multimodal large language models (MLLMs) still underperform on visual mathematical problem solving, especially in reliably perceiving and interpreting diagrams. Inspired by human problem-solving, we hypothesize that the ability to extract meaningful information from diagrams is pivotal, as it directly conditions subsequent inference. Hence, we introduce FlowVerse, a comprehensive benchmark that provides a fine-grained evaluation of MLLMs' perception and reasoning capabilities. Our preliminary results on FlowVerse reveal that existing MLLMs exhibit substantial limitations when extracting essential information and reasoned properties from diagrams and performing complex reasoning based on these visual inputs. In response, we introduce MathFlow, a modular problem-solving pipeline that decouples perception and inference into distinct stages, thereby optimizing each independently. Given the perceptual limitations observed in current MLLMs, we trained MathFlow-P-7B as a dedicated perception model. Experimental results indicate that MathFlow-P-7B yields substantial performance gains when integrated with various closed-source and open-source inference models. This demonstrates the effectiveness of the MathFlow pipeline and its compatibility with diverse inference frameworks. Project page: https://github.com/MathFlow-zju/MathFlow.

This paper studies token issuance strategies in Initial Coin Offering (ICO) markets and examines how an external funding constraint designed to mitigate post-financing fund misappropriation risk affects quality signaling. We develop a three-stage model of fundraising, production, and market trading, in which a venture privately observes project quality and uses token issuance as both a financing instrument and a quality signal. After fundraising, the external funding constraint restricts the use of raised capital and affects subsequent production and token valuation. We show that the external funding constraint has a dual effect. While it reduces fund misappropriation risk by limiting the venture's discretion over capital use, it may weaken quality signaling by increasing the signaling cost of the high-quality venture and reducing the cost for the lowquality venture to imitate the high-quality venture. Furthermore, unlike the conventional signaling intuition in ICO literature that a high-quality venture separates itself by reducing token issuance, we find that the high-quality venture may increase token issuance to signal quality under certain conditions. The optimal issuance strategy depends on investor liquidity structure and the extent of information asymmetry regarding project quality. These results provide guidance for designing optimal token issuance strategies and reveal an important tradeoff in ICO governance: funding constraints designed to reduce post-financing moral hazard may simultaneously affect the effectiveness of quality signaling during fundraising.

Human-machine shared driving (HMSD) has emerged as a crucial transitional paradigm before the widespread adoption of fully autonomous vehicles. However, existing research typically only considers either human-dominated or human-machine equal relationships, neglecting the fact that these two relationships alternate during driving, which leads to a gap between theory and reality. To address this issue, this study proposes a large language model (LLM)-based game equilibrium selection approach for human-machine shared driving authority allocation. Firstly, a game equilibrium selection model is developed to seamlessly transition between Stackelberg equilibrium and Nash equilibrium, addressing human-dominated and human-machine equal relationships, respectively. The selection process is implemented using an LLM, which bases its decisions on scenario understanding. To enhance the LLM’s scenario understanding performance, a set of indicators capturing human-machine conflicts, driver involvement, and collision risks is introduced as prior knowledge. Furthermore, an LLM-based scenario-understanding module is designed to embed knowledge into the LLM and enable it to function effectively within the HMSD system. Finally, a human-in-the-loop experiment is conducted to validate the proposed strategy. The results show that LLMs can understand the provided knowledge, flexibly adapt to different scenarios, and accurately grasp human-machine interactions. Moreover, the proposed strategy effectively reduces human-machine conflicts, better satisfies driver intentions, and reduces driver workload, showcasing the potential of LLM-based decision-making in human-machine interaction.

Data-driven decision-making plays an increasingly important role in engineering management and complex operational systems under uncertainty and dynamic environments. This article reviews the major paradigms in data-driven optimization, including offline learning and stochastic optimization, robust and distributionally robust optimization under small-data regimes, and adaptive online and reinforcement learning approaches. We examine the methodological foundations of these paradigms and discuss their applications in engineering management contexts. Finally, we highlight emerging research directions at the intersection of artificial intelligence and decision-making.

ABSTRACTWe consider the dynamic pricing problem of a monopolist seller who sells a set of mutually substitutable products over a finite time horizon. Customer demand is sensitive to the price of each individual product and the reference price which is formed from a comparison among the prices of all products. To maximize the total expected profit, the seller needs to determine the selling price of each product and also select a reference product (to be displayed) that affects the consumer's reference price. However, the seller initially knows neither the demand function nor the optimal reference product, but can learn them from past observations on the fly. As such, the seller faces the classical trade‐off between exploration (learning the demand function and reference price) and exploitation (using what has been learned thus far to maximize revenue). We propose a rate‐optimal dynamic learning‐and‐pricing algorithm that integrates iterative least squares estimation and bandit control techniques in a seamless fashion. We show that the cumulative regret, that is, the expected revenue loss caused by not using the optimal policy over periods, is upper bounded by where hides any logarithmic factors. We also establish the regret lower bound (for any learning policies) to be . We then generalize our analysis to a more general demand model. Our algorithm performs consistently well numerically, outperforming an exploration‐exploitation benchmark. The use of price experimentation and estimation techniques could be readily applied in real retail management.

PURPOSE:The aim of this study was to analyze adverse events in terms of safety signals and conduct pairwise comparisons on the constituent ratios of the reporting rates, severity, and outcomes of peripheral neuropathy among poly ADP-ribose polymerase inhibitors in the treatment of epithelial ovarian cancer, fallopian tube cancer, and primary peritoneal cancer (collectively referred to as EOC) leveraging the US Food and Drug Administration Adverse Event Reporting System. METHODS:Data on peripheral neuropathy reports related to EOC treatment submitted to the US Food and Drug Administration Adverse Event Reporting System from the first quarter of 2015 to the third quarter of 2024 were collected. Three poly ADP-ribose polymerase inhibitors are identified: olaparib, niraparib, and rucaparib. The primary composite end point of this study was the safety signals for peripheral neuropathy in patients with EOC receiving poly ADP-ribose polymerase inhibitors treatment, whereas the secondary end points included the safety signals for sensory neuropathy, autonomic neuropathy, and motor neuropathy. All analyses were conducted using Stata 18.0 MP software. FINDINGS:A total of 300,810 eligible records were included, among which there were 70,332 reports of peripheral neuropathy. For the primary composite end point, a safety signal related to peripheral neuropathy was detected with niraparib (reporting odds ratio [ROR] = 1.47; information component [IC]025 = 0.21), whereas no safety signal was found with olaparib or rucaparib. For the secondary end points, safety signals related to autonomic, sensory, and motor neuropathies were detected with niraparib (ROR = 1.42, IC025 = 0.21; ROR = 1.39, IC025 = 0.20; ROR = 1.31, IC025 = 0.17), whereas no signals were identified with olaparib and rucaparib. IMPLICATIONS:For patients with EOC, prudent surveillance of peripheral neuropathy is warranted when administrating niraparib. Certainly, more large-scale and long-term follow-up period studies were entailed.

Although Seasonal Influenza Vaccination (SIV) is a crucial preventive measure, achieving sufficient coverage to completely control influenza epidemics poses a significant challenge. This study aims to evaluate optimal strategies for SIV to prevent high-intensity level of influenza epidemics in Zhejiang Province, China. High-intensity outbreaks were defined as weekly incidence rates above 72.2 per 100,000. This study estimated the incidence of influenza from 2018 to 2023 in Zhejiang Province, China, using influenza weekly surveillance data. We developed a Susceptible-Vaccinated-Infectious-Recovered-Susceptible (SVIRS) model to simulate influenza transmission and used a decision tree to assess seven vaccination strategies aimed at preventing high-intensity influenza outbreaks. These strategies differed in vaccine coverage across the three age groups: 0–14 years, 15–59 years, and 60 + years, despite having the same overall vaccination coverage. Between 2018 and 2020, influenza incidence in Zhejiang Province followed typical seasonal patterns. However, during the COVID-19 pandemic, these patterns became irregular, culminating in a high-intensity influenza season in 2022–2023. Model simulations indicated that increasing population-wide vaccination coverage to 36.17

We consider a variant of the classic newsvendor problem in which the firms face both demand and yield randomness. Different from the existing literature, we assume that decision-makers have no priori knowledge of the distribution functions of demand and yield, but have access to past observations of demand, yield, and related feature information. We integrate predictive machine learning algorithms to determine the optimal order quantity directly from historical data, respectively based on the empirical risk minimization (ERM) principle, kernel regression approach, K-nearest neighbors (kNN), and classification and regression trees (CART). These data-driven approaches can not only sufficiently capture useful information from relevant features, but also take into account the structure of the optimization problem, which can effectively avoid inconsistency solutions in the traditional “prediction-then-optimization” approach. Most importantly, we establish out-of-sample generalization error bounds under mild conditions using uniform stability-based and Rademacher complexity-based methods in computational learning theory and then show the asymptotic optimality of the data-driven approaches based on kernel regression and kNN. Our data-driven approaches can tractably deal with both independent and interdependent demand and yield uncertainties. Finally, numerical experiments based on both synthetic data and real data are conducted to compare our proposed methods with two traditional benchmark approaches, including the Sample Average Approximation (SAA) approach and the traditional “Predict-then-Optimize” framework based on CART. We observe that our data-driven approaches can achieve significant performance improvement and the one based on the kernel regression method tends to perform the best on real data, with an average daily cost saving of up tp 54.92%.

We investigate the interaction between cash hedging and responsive pricing in mitigating cash flow risks under price competition. Cash hedging smooths cash flows with financial instruments, while responsive pricing provides firms with operational flexibility to postpone pricing after the realization of cash flows. The firms' cash flows are correlated. We first investigate the equilibria in a single-strategy game where cash hedging and responsive pricing cannot be used simultaneously. Our results show that fierce competition induces heterogeneous choices in cash hedging and responsive pricing, and this heterogeneity is amplified as cash flow correlation increases from negative to positive. We further explore the equilibria in a full game setting where cash hedging and responsive pricing can be used simultaneously. Homogeneous choices within a combined strategy often conflict with firms' profit objectives, as limited pricing flexibility and constrained cost-reduction stability fail to effectively manage cash flow risks. Conversely, heterogeneous choices in a single strategy and a combined strategy can be advantageous, particularly in relatively stable markets.

The proliferation of counterfeit products poses a substantial threat to numerous industries. Blockchain technology (BCT) offers an effective solution for product traceability, providing a means to combat counterfeiting. However, BCT can verify the authenticity of the information but cannot confirm the veracity of the product itself, a problem known as counterfeiting at the source. To our knowledge, this issue has yet to be studied. The security level of BCT traceability is used to indicate its ability to combat counterfeiting. We establish game-theoretical models to investigate BCT adoption strategies for a typically authentic firm and a premium firm to fight counterfeiting in a vertically differentiated competition. This study demonstrates that BCT reduces deceptive counterfeiters’ incentive to pool with the branded firm and mitigates the negative impact of asymmetric information on the prices, market share, and profits of authentic products in a monopoly. In instances where the proportion of counterfeits is substantial, premium products will lose market share, a phenomenon often referred to as “bad money driving out good money.” In a vertically differentiated competition, if the quality of the premium product is below a certain threshold, it is recommended that the premium firm be the first to adopt BCT, while the typically authentic firm should not follow (Scenario NB). That is, Scenario NB is a win-win situation for both firms in the competition. The premium firm that has adopted BCT can offer a “free ride” to the typically authentic firm.

Due to the increase in global energy consumption and carbon dioxide emissions, new energy are eagerly expected to be widely put into application. The installation of photovoltaic systems will increase the fluctuation and uncertainty of load, which will have a certain impact on the planning of customer-side energy storage systems (ESSs). The current tariff policy and the scenarios of energy storage devices have made strict requirements on the ability to control load. In this paper, an energy storage revenue assessment method based on portfolio theory is proposed. The uncertainty of load is analysed by non-parametric kernel density estimation (KDE). Then a factor is used to measure the risk of reducing the load, which is coped with the ESS. The capacity and operation strategies of ESS is optimally allocated by linear programming (LP). Then a method based on portfolio theory is used to quantify the combined impact of risk and revenue, which provides a certain reference for the determination of the capacity of the behind-the-meter (BTM) ESS and the operation strategy.

E-commerce live streaming has emerged as a novel electronic shopping method, progressively becoming a pivotal force in the transformation of the retail industry and driving consumer growth and economic prosperity. Compared to traditional TV shopping channels, live e-commerce platforms provide celebrities with significant autonomy in their marketing strategies. This study develops a marketing model to analyze new product promotions through live-streaming collaborations between brand owners and celebrities, focusing on balancing economic gains with the sustainability of the following base. The findings reveal that the evaluation of the product-market fit of a celebrity and their professional competence significantly influence the promotional strategies. Highly professional celebrities favor honest recommendations for high-fit products, while deceptive promotions mitigate risks for low-fit products. Moderately professional celebrities adopt nuanced strategies, shifting between honest and deceptive promotions based on product-market fit levels, while low-professionalism celebrities consistently favor deceptive promotions due to inaccurate product judgment. Furthermore, the coefficient of follower traffic value and follower concentration shape the effectiveness of the strategy, forming customized approaches for celebrities with short-sighted, mature, and new arrivals. These insights optimize marketing strategies by aligning them with product traits, audience composition, and engagement dynamics.

To compare the augmentative efficacy of second-generation anti-psychotics (SGA) to anti-depressants in adult patients with treatment-resistant depression (TRD) adjusting follow-up period and explore the underlying"time window"effects of the regimens. Databases included Embase, PubMed, Scopus, Cochrane Library and Google Scholars as well as Clinicaltrials.gov from inception to May 15, 2024, for relevant randomized controlled studies (RCTs) were retrieved. The primary endpoint was Montgomery Asberg Depression Rating Scale (MADRS). The secondary endpoint was MADRS response rate. The tertiary endpoints were Clinical Global Impression-severity (CGI-S) and MADRS remission rate. Standard mean difference (SMD) and hazard ratio (HR) were generated by Bayesian network meta-regression (NMR) for pairwise comparisons on dichotomous and consecutive variants, respectively. A total of 23 studies (N = 10679) with 24 augmentation agents were included in the NMR. For the primary endpoint, compared with ADT, aripiprazole 3 - 12 mg/d, brexpiprazole 1 - 3 mg/d, cariprazine 1.5 - 3 mg/d, olanzapine 6 - 12 mg/d and fluoxetine 25 - 50 mg/d combination, and quetiapine XR were significantly effective (SMD ranged from - 0.28 to - 0.114) and their effect sizes were comparable, after adjusting follow-up period, the results resembled the former except for quetiapine XR (SMD = - 0.10, 95

Energy flexibility is an important way to realize the real-time balance of regional energy supply and demand,and it can also reduce the operating cost of the system combined with peak-valley electricity price.The operation of central hot water system in university dormitories can provide abundant energy flexibility due to the heat storage characteristics of water tank and pipe network.In order to quantify the potential of the flexibility of the central hot water system in university dormitories to optimize the operation of the system,an optimization model of the central hot water system based on model predictive control was proposed in this study.The proposed model consists of the water consumption prediction model based on the circulation neural network,the heat source model based on the linear model,and the hot water system temperature prediction model based on the RC(resistance and the capacitance)model.Putting emphasis on the reduction of operating cost as the objective function,genetic algorithm is used to find the optimal operation strategy.The results obtained from simulation showed that using energy flexibility to transfer hot water load from peak electricity price period to valley electricity price period could save 30%and 9.4%operating costs in summer and winter,respectively,and increase 64.4%power consumption during valley electricity price period in typical winter months,while bringing more suitable hot water temperature and comfort.

This study addresses the challenge of optimizing control strategies for multi-energy systems in high-density residential buildings, focusing on space heating and domestic hot water applications in cold climates. Due to constrained space, dispersed energy consumption patterns, and limited renewable energy integration in such buildings, a multi-energy system combining solar thermal collectors, air source heat pumps, and sewage source heat pumps with an embedded stratified thermal storage tank is proposed. A novel quality-quantity regulation strategy for the demand side dynamically adjusts both the circulating flow rate and temperature of the heating fluid based on tank energy status and load requirements, enhancing energy utilization and extending heat pump operation compared to traditional constant-flow quality regulation. For the supply side, dead-zone control strategies are optimized for each subsystem: solar collectors (upper/lower dead-band temperature difference), air source heat pumps (control based on supply-required temperature difference), and other. Simulation results demonstrate that the quality-quantity strategy better matches heating loads, improves tank stratification, and increases heat pump runtime. Solar subsystem performance is highly sensitive to the upper dead-band limit, while sewage source heat pumps control has minimal impact on solar operation. The findings provide practical guidance for multi-energy system control design, improving efficiency and reducing auxiliary heating dependence in residential buildings.

Many digital platforms provide a search environment for consumers to evaluate sellers' products. We investigate a strategic platform's preference over two classical search patterns-parallel versus sequential-keeping in check consumers' search behavior (how many products and attributes to evaluate) and sellers' strategies (price and assortment decisions). In the benchmark model with exogenous assortment level, our results show that the platform prefers a parallel (sequential) pattern when the search cost is small (large) or when the assortment level is high (low). However, when the assortment level is a decision by sellers, the platform's preference will be altered qualitatively: The platform prefers a parallel (sequential) pattern when the search cost is large (small), and the analytical predictions are generally consistent with observations in practice. We have identified several novel effects that are built on the fundamental difference between parallel and sequential patterns and use them to explain the platform's search-pattern preference. Interestingly, our paper shows that the platform can strategically use operational means (assortment prevention effect) and marketing means (pricing prevention effect) to manipulate consumers' search to maximize its profit.