Agent are structured collections of instructions, tools, and supporting resources that help large language model (LLM) agents perform particular classes of tasks. Empirical evidence shows that the design of can materially affect agent task performance, yet systematically optimizing remains challenging. Since a comprises instructions, tools, and supporting resources in a structured way, optimizing it requires jointly determining both the structure of these components and the content each component contains. This gives rise to a complex decision space with strong interdependence across structure and components. We therefore represent these two coupled decisions as structure and component content, and formulate optimization as a bilevel optimization problem. We propose a bilevel optimization framework in which an outer loop employs Monte Carlo Tree Search to determine the structure, while an inner loop refines the component content within the structure selected by the outer loop. In both loops, we employ LLMs to assist the optimization procedure. We evaluate the proposed framework on an open-source Operations Research Question Answering dataset, and the experimental results suggest that the bilevel optimization framework improves the performance of the agents with the optimized .

Transportation-enabled service (TRENS) platforms, including ride-hailing, food and grocery delivery, and crowdsourced logistics, are platform markets in which pricing, matching, routing, service recovery, and information design continuously reshape the behavior of customers, drivers, couriers, merchants, and operators. This survey asks where large language model (LLM) agents can credibly improve such systems. The opportunity, we argue, is not replacing mature operations research (OR), machine learning (ML), or reinforcement learning (RL) pipelines, but two interfaces those pipelines handle poorly. First, behavioral agents can enrich offline policy evaluation by simulating heterogeneous stakeholder responses to changed prices, incentives, service promises, and information structures. Second, operational agents can translate unstructured language, documents, regulations, complaints, and managerial intent into structured inputs for platform workflows and optimization systems. Reviewing classical TRENS baselines, LLM behavioral- and operational-agent studies, and industrial systems from Meituan, DiDi, Instacart, Uber Freight, and DoorDash, we find a consistent pattern: current production evidence is concentrated in semantic-heavy workflow layers such as customer service, catalog/query understanding, document processing, internal analytics, and passenger or logistics copilots, while public evidence of LLMs deployed directly on the critical operational path for real-time dispatching or pricing remains limited. The resulting research agenda prioritizes revealed-preference calibration of behavioral agents, bilevel models coupling platform controls with simulated stakeholder response, explicit latency and verification contracts for hybrid systems, and TRENS-specific benchmarks that jointly evaluate behavior, optimization, and workflow automation.

Building a blockchain digital twin largely requires translating domain knowledge and specific system descriptions into a simulator architecture, calibrating its parameters against behavioral evidence, and validating the constructed twin. These steps are commonly performed through application-specific modeling efforts that can be difficult to reuse across systems and downstream decision problems. We consider automating this process through Spec2Twin-Chain, a framework that formulates blockchain digital-twin construction as a bi-level optimization problem. At the upper level, a large language model proposes and revises structurally admissible architectures using system specifications, behavioral evidence, and feedback from evaluated designs. At the lower level, a simulation-based optimizer calibrates the architecture-conditioned parameters under explicit objectives and guardrail constraints. The two levels iterate. The evaluated candidates at lower levels are retained in a global archive and used to guide subsequent proposals at upper levels. We conduct controlled experiments involving twin calibration, feedback-driven recovery, stress analysis, downstream policy optimization, and policy updating. The results demonstrate that the framework can construct behaviorally accurate twins, improve initial designs through iterative feedback, and reuse calibrated twins to support downstream decisions.

The software and data in this repository are a snapshot of the software and data that were used in the research reported on in the paper Machine Learning-Assisted Stochastic Kriging for Offline Simulation Online Application by Haoting Zhang, Jinghai He, Rhonda Righter, Zuo-Jun Max Shen, Zeyu Zheng. The snapshot is based on this SHA in the development repository.

Short-video platforms are closed-loop, human-in-the-loop ecosystems where platform policy, creator incentives, and user behavior co-evolve. This feedback structure makes counterfactual policy evaluation difficult in production, especially for long-horizon and distributional outcomes. The challenge is amplified as platforms deploy AI tools that change what content enters the system, how agents adapt, and how the platform operates. We propose a large language model (LLM)-augmented digital twin for short-video platforms, with a modular four-twin architecture (User, Content, Interaction, Platform) and an event-driven execution layer that supports reproducible experimentation. Platform policies are implemented as pluggable components within the Platform Twin, and LLMs are integrated as optional, schema-constrained decision services (e.g., persona generation, content captioning, campaign planning, trend prediction) that are routed through a unified optimizer. This design enables scalable simulations that preserve closed-loop dynamics while allowing selective LLM adoption, enabling the study of platform policies, including AI-enabled policies, under realistic feedback and constraints.

For many real-time decision problems in complex stochastic systems, a decision maker observes the system status online in real time and needs to immediately evaluate the performance of alternative decision choices based on the observed system status. Simulation metamodels can effectively support these needs by constructing a fast-to-evaluate mapping from the system status and decision to system performance using offline-generated simulation samples. However, when the system status involves high-dimensional information and a large number of simulated samples, classical simulation metamodeling approaches, such as stochastic kriging, may face challenges in terms of model specification, computational complexities, and computer storage demands. To address these challenges, we propose using machine learning models to assist stochastic kriging in building metamodels offline, which then can support online applications. The machine learning models can capture the potential nonlinear dependence of the stochastic kriging parameters on the high-dimensional system status. We analyze standard properties, such as mean squared errors and uncertainty quantification, for the proposed machine learning-assisted metamodel, and we show its consistency and asymptotic validity. We demonstrate the comparative advantage of our approach through numerical experiments. History: Accepted by Bruno Tuffin, Area Editor for Simulation. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2023.0130 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2023.0130 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

The field of simulation optimization (SO) encompasses various methods developed to optimize complex, expensive-to-sample stochastic systems. Established methods include, but are not limited to, ranking-and-selection for finite alternatives and surrogate-based methods for continuous domains, with broad applications in engineering and operations management. The recent advent of large language models (LLMs) offers a new paradigm for exploiting system structure and automating the strategic selection and composition of these established SO methods into a tailored optimization procedure. This work introduces SOCRATES (Simulation Optimization with Correlated Replicas and Adaptive Trajectory Evaluations), a novel two-stage procedure that leverages LLMs to automate the design of tailored SO algorithms. The first stage constructs an ensemble of digital replicas of the real system. An LLM is employed to implement causal discovery from a textual description of the system, generating a structural `skeleton' that guides the sample-efficient learning of the replicas. In the second stage, this replica ensemble is used as an inexpensive testbed to evaluate a set of baseline SO algorithms. An LLM then acts as a meta-optimizer, analyzing the performance trajectories of these algorithms to iteratively revise and compose a final, hybrid optimization schedule. This schedule is designed to be adaptive, with the ability to be updated during the final execution on the real system when the optimization performance deviates from expectations. By integrating LLM-driven reasoning with LLM-assisted trajectory-aware meta-optimization, SOCRATES creates an effective and sample-efficient solution for complex SO optimization problems.

We study the problem of quantifying simulation uncertainty when the input data to the simulation model are contaminated by and mixed with outliers. We first propose an input influence function to quantify the sensitivity of the simulation model to the input distribution perturbed by arbitrary data. Then we provide a “robustify-then-quantify” approach to evaluate input uncertainty in stochastic simulation, based on the robust estimation for the input model. We show asymptotic consistency for the proposed procedure when the proportion of outliers tends to zero. For settings with general proportions of outliers, we illustrate through two numerical experiments that the proposed procedure has stable performances in terms of the cover ratio and confidence interval length.

Motivated by the growing need for black-box optimization and data privacy, we introduce a collaborative Bayesian optimization (BO) framework that addresses both of these challenges. In this framework agents work collaboratively to optimize a function they only have oracle access to. In order to mitigate against communication and privacy constraints, agents are not allowed to share their data but can share their Gaussian process (GP) surrogate models. To enable collaboration under these constraints, we construct a central model to approximate the objective function by leveraging the concept of Wasserstein barycenters of GPs. This central model integrates the shared models without accessing the underlying data. A key aspect of our approach is a collaborative acquisition function that balances exploration and exploitation, allowing for the optimization of decision variables collaboratively in each iteration. We prove that our proposed algorithm is asymptotically consistent and that its implementation via Monte Carlo methods is numerically accurate. Through numerical experiments, we demonstrate that our approach outperforms other baseline collaborative frameworks and is competitive with centralized approaches that do not consider data privacy.

This directory contains the code for the Clustering then Estimation of Spatio-Temporal Self-Exciting Processes (CTE) algorithm.

In healthcare applications, there is a growing need to develop machine learning models that use data from a single source, such as that from a wrist wearable device, to monitor physical activities, assess health risks, and provide immediate health recommendations or interventions. However, the limitation of using single-source data often compromises the model's accuracy, as it fails to capture the full scope of human activities. While a more comprehensive dataset can be gathered in a lab setting using multiple sensors attached to various body parts, this approach is not practical for everyday use due to the impracticality of wearing multiple sensors. To address this challenge, we introduce a transfer learning framework that optimizes machine learning models for everyday applications by leveraging multi-source data collected in a laboratory setting. We introduce a novel metric to leverage the inherent relationship between these multiple data sources, as they are all paired to capture aspects of the same physical activity. Through numerical experiments, our framework outperforms existing methods in classification accuracy and robustness to noise, offering a promising avenue for the enhancement of daily activity monitoring.

The proliferation of language models has marked a significant advancement in technology and industry in recent years. The training of these models largely involves human feedback, a procedure that faces challenges including intensive resource demands and subjective human preferences. In this work, we incorporate feedback provided by artificial intelligence (AI) models instead of relying entirely on human feedback. We propose a simulation optimization framework to train the language model. The objective function for training is approximated using feedback from both human and AI models. We employ the method of control variate to reduce the variance of the approximated objective function. Additionally, we provide a procedure for deciding the sample size to acquire preferences from both human and AI models. Numerical experiments demonstrate that our proposed procedure enhances the performance of the language model.

We propose a new estimation procedure for general spatio-temporal point processes that include a self-exciting feature. Estimating spatio-temporal self-exciting point processes with observed data is challenging, partly because of the difficulty in computing and optimizing the likelihood function. To circumvent this challenge, we employ a Poisson cluster representation for spatio-temporal self-exciting point processes to simplify the likelihood function and develop a new estimation procedure called “clustering-then-estimation” (CTE), which integrates clustering algorithms with likelihood-based estimation methods. Compared with the widely used expectation-maximization (EM) method, our approach separates the cluster structure inference of the data from the model selection. This has the benefit of reducing the risk of model misspecification. Our approach is computationally more efficient because it does not need to recursively solve optimization problems, which would be needed for EM. We also present asymptotic statistical results for our approach as theoretical support. Experimental results on several synthetic and real data sets illustrate the effectiveness of the proposed CTE procedure. History: Accepted by Ram Ramesh, Area Editor for Data Science & Machine Learning. Funding: J. Anderson is supported by NSF [Grant ECCS-2144634]. R. Righter is supported by the Ron Wolff Chaired Professorship. Z. Zheng is supported by NSF [Grant DMS-2220537]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.0351 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2022.0351 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

Contextual decision-making problems have witnessed extensive applications in various fields such as online content recommendation, personalized healthcare, and autonomous vehicles, where a core practical challenge is to select a suitable surrogate model for capturing unknown complicated reward functions. It is often the case that both high approximation accuracy and explicit uncertainty quantification are desired. In this work, we propose a neural network-accompanied Gaussian process (NN-AGP) model, which leverages neural networks to approximate the unknown and potentially complicated reward function regarding the contextual variable, and maintains a Gaussian process surrogate model with respect to the decision variable. Our model is shown to outperform existing approaches by offering better approximation accuracy thanks to the use of neural networks and possessing explicit uncertainty quantification from the Gaussian process. We also analyze the maximum information gain of the NN-AGP model and prove regret bounds for the corresponding algorithms. Moreover, we conduct experiments on both synthetic and practical problems, illustrating the effectiveness of our approach.

We study the problem of simulating a class of nonstationary spatio-temporal Poisson processes. The Poisson intensity function is non-stationary and piecewise linear in both the time dimension and the spatial location dimensions. We propose an exact simulation algorithm based on the inversion method. This simulation algorithm adopts three advantages. First, the entire procedure involves only closed-form computation with no need for numerical integration or numerical inversion of any function. Each step in the algorithm only requires exact arithmetic operations. Second, the proposed algorithm is sample efficient, especially compared to the thinning method when the maximum intensity value is much larger than the minimum intensity value. Third, the algorithm generates arrivals sequentially, one at a time in ascending order, so that they can be conveniently fed into real-time or online decision-making tools.

This is a supplement document for the WSC2020 paper con342s1.

Time Series Classification (TSC) has been an important and challenging task in data mining, especially on multivariate time series and multi-view time series data sets. Meanwhile, transfer learning has been widely applied in computer vision and natural language processing applications to improve deep neural network's generalization capabilities. However, very few previous works applied transfer learning framework to time series mining problems. Particularly, the technique of measuring similarities between source domain and target domain based on dynamic representation such as density estimation with importance sampling has never been combined with transfer learning framework. In this paper, we first proposed a general adaptive transfer learning framework for multi-view time series data, which shows strong ability in storing inter-view importance value in the process of knowledge transfer. Next, we represented inter-view importance through some time series similarity measurements and approximated the posterior distribution in latent space for the importance sampling via density estimation techniques. We then computed the matrix norm of sampled importance value, which controls the degree of knowledge transfer in pre-training process. We further evaluated our work, applied it to many other time series classification tasks, and observed that our architecture maintained desirable generalization ability. Finally, we concluded that our framework could be adapted with deep learning techniques to receive significant model performance improvements.