Showing papers by "Alexander G. Nikolaev published in 2019"

Adversarial Balancing-based Representation Learning for Causal Effect Inference with Observational Data

Abstract: Learning causal effects from observational data greatly benefits a variety of domains such as health care, education and sociology. For instance, one could estimate the impact of a new drug on specific individuals to assist the clinic plan and improve the survival rate. In this paper, we focus on studying the problem of estimating Conditional Average Treatment Effect (CATE) from observational data. The challenges for this problem are two-fold: on the one hand, we have to derive a causal estimator to estimate the causal quantity from observational data, where there exists confounding bias; on the other hand, we have to deal with the identification of CATE when the distribution of covariates in treatment and control groups are imbalanced. To overcome these challenges, we propose a neural network framework called Adversarial Balancing-based representation learning for Causal Effect Inference (ABCEI), based on the recent advances in representation learning. To ensure the identification of CATE, ABCEI uses adversarial learning to balance the distributions of covariates in treatment and control groups in the latent representation space, without any assumption on the form of the treatment selection/assignment function. In addition, during the representation learning and balancing process, highly predictive information from the original covariate space might be lost. ABCEI can tackle this information loss problem by preserving useful information for predicting causal effects under the regularization of a mutual information estimator. The experimental results show that ABCEI is robust against treatment selection bias, and matches/outperforms the state-of-the-art approaches. Our experiments show promising results on several datasets, representing different health care domains among others.

Monitoring Multimodal Travel Environment Using Automated Fare Collection Data: Data Processing and Reliability Analysis

Abstract: Monitoring transit system “health” by extracting and tracking such quantities as travel time, transfer time, number of passengers, etc., is critical to the benefit of travelers, planners and operators within a transit system. Most of the data typically available to and useful for analysts are generated by tracking vehicles instead of individual passengers/travelers—these data are useful, albeit within certain limits. This paper presents methods for obtaining system-level transit information from a new type of data—that coming from an Automated Fare Collection (AFC) system,—which provides hour-to-hour, day-to-day transit information, such as the value and reliability in both travel time and traveler count, and the location of congested road clusters in a city. The AFC data of public transit system in Seoul, South Korea is used as an example to illustrate the proposed data extraction methods and analysis. This paper is structured and detailed so as to provide both methodological and practical guidance for researchers and data-handling analysts.

Exploring Social Media Network Landscape of Post-Soviet Space

Abstract: The “post-Soviet space” consists of countries with a substantial fraction of the world’s population; however, unlike many other regions, its social media network landscape is still somewhat under-explored. This paper aims at filling this gap. To this purpose, we use anonymized data on user friendships at VK.com (also known as VKontakte and, informally, as “Russian Facebook”), which is the largest and most popular social media portal in the post-Soviet space with hundreds of millions of user accounts. Using the VK network snapshots from October 2015 to December 2016, we conduct a “multiscale” empirical study of this network by considering connections among individual users, cities, and countries. Our findings indicate that the VK users form a small-world network with basic characteristics consistent with Facebook and other social media networks. In addition, the analysis of modularity-based communities within the user scale network reveals a pattern of geographical separation of the identified communities mostly along the borders between countries. However, the comparison of the two network snapshots suggests that some of these communities may be “blending” within the network, whereas other communities remain “self-contained.” Furthermore, the analysis of city scale and country scale networks identifies cities and countries that are most “central” (in the context of certain metrics) in the VK network.

Causal analysis of the impact of homecare services on patient discharge disposition.

Abstract: This study uses observational causal inference to evaluate the impact of different combinations of home care services (nursing, therapies, social work, home aides) on end-of-episode disposition for individuals with chronic diseases associated with the circulatory, endocrine, and musculoskeletal systems. The potential to generate actionable recommendations for personalizing home care services, or treatment plans, from limited clinical and care needs data is demonstrated. For patients with chronic disease in the circulatory or musculoskeletal systems, a 2.91% and 3.38% decrease, respectively, in acute care hospitalization rates could be obtained by providing patients with therapy and nursing services, rather than therapy services alone.

Robust communication network formation: a decentralized approach

Abstract: The formation of robust communication networks between independently acting agents is of practical interest in multiple domains, for example, in sensor placement and Unmanned Aerial Vehicle communication. These are the cases where it is only feasible to have the communicating actors modify the network locally, i.e., without relying on the knowledge of the entire network structure and the other actors’ decisions. This calls for approaches to optimizing network structure in a decentralized way. We present an actor-oriented modeling approach to design and parameterize models that enable the creation of networks that exhibit the properties desirable for efficient information sharing. Computational experiments show that the achieved network formation rules, specified in a calculated way, allow agents to maintain robust network structure by activating only a limited number of direct communication channels. The obtained results are promising, as evidenced by the reported comparisons to optimal network configuration solutions obtained in a centralized way.

Sampled Fictitious Play on Networks

Abstract: We formulate and solve the problem of optimizing the structure of an information propagation network between multiple agents. In a given space of interests (e.g., information on certain targets), each agent is defined by a vector of their desirable information, called filter, and a vector of available information, called source. The agents seek to build a directed network that maximizes the value of the desirable source-information that reaches each agent having been filtered en route, less the expense that each agent incurs in filtering any information of no interest to them. We frame this optimization problem as a game of common interest, where the Nash equilibria can be attained as limit points of Sampled Fictitious Play (SFP), offering a method that turns out computationally effective in traversing the huge space of feasible networks on a given node set. Our key idea lies in the creative use of history in SFP, leading to the new History Value-Weighted SFP method. To our knowledge, this is the first successful application of FP for network structure optimization. The appeal of our work is supported by the outcomes of the computational experiments that compare the performance of several algorithms in two settings: centralized (full information) and decentralized (local information).