-
作者：Wen, Xin; Sun, Will Wei; Zhang, Yichen
作者单位：New York University; Purdue University System; Purdue University
摘要：Contemporary applications, such as recommendation systems and mobile health monitoring, require real-time processing and analysis of sequentially arriving highdimensional tensor data. Traditional offline learning, involving the storage and utilization of all data in each computational iteration, becomes impractical for these tasks. Furthermore, existing low-rank tensor methods lack the capability for online statistical inference, which is essential for real-time predictions and informed decisi...
-
作者：Yue, Man-Chung; Rychener, Yves; Kuhn, Daniel; Nguyen, Viet Anh
作者单位：University of Hong Kong; Swiss Federal Institutes of Technology Domain; Ecole Polytechnique Federale de Lausanne
摘要：The state-of-the-art methods for estimating high-dimensional covariance matrices all shrink the eigenvalues of the sample covariance matrix toward a data-insensitive shrinkage target. The underlying shrinkage transformation is either chosen heuristically-without compelling theoretical justification-or optimally in view of restrictive distributional assumptions. In this paper, we propose a principled approach to construct covariance estimators without imposing restrictive assumptions. That is, ...
-
作者：Qu, Zhaonan; Galichon, Alfred; Gao, Wenzhi; Ugander, Johan
作者单位：New Jersey Institute of Technology; Columbia University; New York University; New York University; Institut d'Etudes Politiques Paris (Sciences Po); Stanford University; Stanford University
摘要：For a broad class of models widely used in practice for choice and ranking data based on the Luce choice axiom, including the Bradley-Terry-Luce and Plackett-Luce models, we show that the associated maximum likelihood estimation problems are equivalent to a classic matrix-balancing problem with target row and column sums. This perspective opens doors between two seemingly unrelated research areas and allows us to unify existing algorithms in the choice-modeling literature as special instances ...
-
作者：Correa, Jose; Cristi, Andres; Norouzi-Fard, Ashkan; Norouzi-Fard, Ashkan
作者单位：Universidad de Chile; Alphabet Inc.; Google Incorporated
摘要：There is growing awareness and concern about fairness in machine learning and algorithm design. This is particularly true in online selection problems, where decisions are often biased: for example, when assessing credit risks or hiring staff. We address the issues of fairness and bias in online selection by studying multicolor versions of the classic secretary and prophet problems. In the multicolor secretary problem, we consider that each candidate has a color, and we can only compare candid...
-
作者：Dupret, Jean-Loup; Hainaut, Donatien
作者单位：Swiss Federal Institutes of Technology Domain; ETH Zurich; Universite Catholique Louvain
摘要：This paper introduces the generalized policy iteration physics-informed neural network algorithm, a novel numerical scheme for solving continuous-time stochastic optimal control problems in high dimensions when the optimal control does not admit an explicit solution. The proposed iterative deep learning algorithm leverages physicsinformed neural networks on the Hamilton-Jacobi-Bellman residuals in an actor-critic fashion, which is built from the generalized policy iteration technique. It emplo...
-
作者：Yuan, Quan; Du, Longyuan; Hu, Ming
作者单位：Zhejiang University; University of San Francisco; University of Toronto
摘要：We study dynamic pricing under a stochastic, nonlinear, self-exciting demand arrival process over a finite sales horizon. We adopt such a correlated demand process to capture the phenomenon that customers who have made a purchase can inform and excite future customers to arrive. Specifically, the stochastic arrival intensity is boosted immediately after any purchase and gradually decays between consecutive purchases. The arrival intensity is also affected by the market phase the seller operate...
-
作者：Abdallah, Tarek; Reed, Josh
作者单位：Northwestern University; New York University
摘要：We study the single-item dynamic pricing problem in three separate asymptotic regimes characterized by their ratio of inventory to market size. We first consider the case of customer item valuations following an exponential distribution, for which we derive a sharp characterization of the boundaries between each of the regimes. We then proceed to the case of customer item valuations following a general distribution. In this case, we derive for each regime approximations to the optimal value fu...
-
作者：Murthy, Yashaswini; Moharrami, Mehrdad; Srikant, Rayadurgam
作者单位：California Institute of Technology; University of Iowa; University of Illinois System; University of Illinois Urbana-Champaign
摘要：Modified policy iteration (MPI) is a dynamic programming algorithm that combines elements of policy iteration and value iteration. The convergence of MPI is wellstudied in the context of discounted and average-cost Markov decision processes (MDPs). In this work, we consider the exponential cost risk-sensitive MDP formulation, which is known to provide some robustness to model parameters. Although policy iteration and value iteration are well-studied in the context of risk-sensitive MDPs, MPI i...
-
作者：Xu, Zhandong; Li, Zhengyang; Xie, Jun; Chen, Anthony; Liu, Xiaobo
作者单位：Southwest Jiaotong University; Hong Kong Polytechnic University
摘要：This study generalizes the single-class traffic assignment problem of Nikolova and Stier-Moses [Nikolova and Stier-Moses (2014) A mean-risk model for the traffic assignment problem with stochastic travel times. Oper. Res. 62(2):366-382] by considering the continuously distributed risk-aversion factor, termed the continuous mean-risk traffic assignment (CMRTA) problem. In CMRTA, travelers categorized into infinitely many user classes play a congestion routing game based on their risk attitude t...
-
作者：Han, Jiangze; Ryan, Christopher Thomas; Tong, Xin T.
作者单位：Columbia University; University of British Columbia; National University of Singapore
摘要：Loot boxes are a primary source of revenue in the video game industry. Loot boxes randomly drop items of differing value. To design a loot box, sellers must choose the loot box's purchase price and drop rate (or drop probability) of each item. We show that, in general, the loot box design problem is NP-hard. By restricting the form of player utilities, we can solve the problem exactly in polynomial time when the number of items is fixed. Under different restrictions, we solve the problem appro...
