Attitude towards information in multi-agent settings: Understanding and mitigating Avoidance and Over-Evaluation

多主体环境中对信息的态度：理解和减轻回避和过度评估

• 批准号: 1919453
• 负责人: Matteo Pozzi
• 金额: $ 50万
• 依托单位: Carnegie-Mellon University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-08-15 至 2024-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1919453&HistoricalAwards=false
• 关键词: Attitude; towards; information; multi; agent

New technologies in sensing, data communication and processing allow for extensive instrumentation of the built environment, and the massive flow of information collectable by sensors can transform the operation and the functionality of urban systems. However, this development depends also on the attitude of citizens and stakeholders toward information. This project investigates how interacting agents take decisions about collecting information, with focus on users and managers of urban systems interacting with public policies. For rational and isolated agents acting without external constraints, "information never hurts" and data with low impact on the agents' belief have a small value. This implies, for example, that these agents are always willing to install free (or cheap) sensors, and to install expensive ones only if they provide high-impact information. However, these intuitive properties do not hold true in multi-agent settings, when agents compete one against each other, nor for agents acting under external constraints as those imposed by regulations. Integrating analysis in social science, engineering and computer science, the project will develop a framework for modeling the attitude towards information in these contexts, depending on the agents' preference and the external regulations.The goals of the project are: 1) To develop a framework for assessing the Value of Information in multi-agent settings, modeling the interaction between policy makers and decision makers following external regulations, 2) to gather and analyze empirical data about the attitude toward information, using surveys and interviews among users, and calibrate the models developed in (1), 3) to design mechanisms alleviating Information Avoidance and Over Evaluation, and assess their effectiveness. The project integrates probabilistic models of quantities to be measured and of sensor performance, agents' utility functions and external constraints, optimization methods and behavior modeling, to assess the Value or Information via Bayesian pre-posterior analysis. Such approach will allow understanding how Information Avoidance and Over Evaluation arise, and how appropriate mechanisms of incentives and regulations can mitigate them. The project's outcomes will be key for a better empirical understanding of the attitude towards information, for developing effective large-scale monitor of the built environment and public policies promoting effective information collection, integrating societal and agents' utilities.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

传感、数据通信和处理方面的新技术允许对建筑环境进行广泛的仪表化，传感器收集的大量信息流可以改变城市系统的运作和功能。然而，这种发展也取决于公民和利益攸关方对信息的态度。该项目研究了互动代理如何收集信息的决策，重点是与公共政策互动的城市系统的用户和管理者。对于没有外部约束的理性和孤立的代理人，“信息永远不会伤害”和对代理人信念影响较小的数据具有较小的价值。这意味着，例如，这些代理人总是愿意安装免费（或便宜）的传感器，并安装昂贵的，只有当他们提供高影响力的信息。然而，这些直观的属性并不适用于多代理设置，当代理相互竞争，也不为代理下的外部约束所施加的法规。该项目将结合社会科学、工程学和计算机科学的分析，开发一个框架，用于根据代理人的偏好和外部规则，在这些背景下模拟对信息的态度。该项目的目标是：1）开发一个框架，用于评估多主体环境中的信息价值，模拟政策制定者和决策者之间遵循外部法规的互动，2）通过对用户的调查和访谈，收集和分析有关信息态度的经验数据，并对（1）中开发的模型进行校准; 3）设计缓解信息回避和过度评价的机制，并评估其有效性。 该项目集成了待测量和传感器性能的概率模型、智能体的效用函数和外部约束、优化方法和行为建模，通过贝叶斯前后分析评估价值或信息。这种方法将有助于理解信息规避和过度评价是如何产生的，以及适当的激励和监管机制如何减轻这些问题。该项目的成果将是更好地从经验上理解对信息的态度，开发有效的大规模建筑环境监测和促进有效信息收集的公共政策，整合社会和代理人的效用的关键。该奖项反映了NSF的法定使命，并被认为值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估来支持。

项目成果

Efficient Algorithms for Learning Revenue-Maximizing Two-Part Tariffs

用于学习收入最大化的两部分关税的有效算法

• DOI: 10.24963/ijcai.2020/47
• 发表时间: 2020
• 期刊: {IJCAI-20}
• 作者: Balcan, Maria-Florina;Prasad, Siddharth;Sandholm, Tuomas
• 通讯作者: Sandholm, Tuomas

Learning to Link

学习链接

• 发表时间: 2020
• 期刊: International Conference on Learning Representation
• 作者: Balcan, Maria-Florina;Dick, Travis;Lang, Manuel
• 通讯作者: Lang, Manuel

How much data is sufficient to learn high-performing algorithms? generalization guarantees for data-driven algorithm design

• DOI: 10.1145/3406325.3451036
• 发表时间: 2021-06
• 期刊: Proceedings of the 53rd Annual ACM SIGACT Symposium on Theory of Computing
• 作者: Maria-Florina Balcan;Dan F. DeBlasio;Travis Dick;Carl Kingsford;T. Sandholm;Ellen Vitercik

Learning Predictions for Algorithms with Predictions

通过预测来学习算法的预测

• 发表时间: 2022
• 期刊: Advances in Neural Information Processing Systems
• 作者: Khodak, Mikhail;Balcan, Maria Florina;Talwalkar, Ameet;Vassilvitskii, Sergei
• 通讯作者: Vassilvitskii, Sergei

The Demand for, and Avoidance of, Information

• DOI: 10.1287/mnsc.2021.4244
• 发表时间: 2021-12-15
• 期刊: MANAGEMENT SCIENCE
• 影响因子: 5.4
• 作者: Golman, Russell;Loewenstein, George;Saccardo, Silvia
• 通讯作者: Saccardo, Silvia