Research

Working Papers

Economics Research

Market Segmentation Under Costly Information Acquisition

This paper investigates third-degree price discrimination under endogenous market segmentation. Segmenting a market requires access to information about consumers, and this information comes with a cost. I explore the trade-offs between the benefits of segmentation and the costs of information acquisition, revealing a non-monotonic relationship between consumer surplus and the cost of information acquisition for monopolist. I show that in some markets, allowing the monopolist easier access to customer data can also benefit customers on average. I also analyzed how social welfare reacts to changes in the cost level of information acquisition and showed that the non-monotonicity result is also valid in social welfare analysis. This result can be considered a good caveat for policymakers who focus on market efficiency. 

You can access the paper here

Advisor: Yuhta Ishii

Empirical Analysis of Optimality of Auction Mechanisms: A Structural Approach


This study employs a structural econometric model to investigate whether real-world auctions align with Myerson's optimal auction design theory in the Independent Private Value (IPV) setting. By analyzing data from repeated auctions, particularly first-price or second-price formats, I aim to estimate the underlying distribution of bidder valuations  F(v) to determine if it exhibits increasing hazard rates, a key component of the IPV framework.

Advisor: Karl Schurter


Computer Science Research

Classification of Dialogue Segment Breaks in GUIDE Dataset

Collaborators:Ming Zhu,Sai Chandana Priya,Lakshmi Chandrika,Yichao Chen 

Dept. of CSE, Pennsylvania State University

Supervisor: PROF. REBECCA J. PASSONNEAU


Abstract

In this paper, we explore the task of classifying conversation segment breaks using various Natural Language Processing (NLP) models. We leverage the rich textual data within the GUIDE dataset to identify these transitions. We have worked with a couple of baseline models alongside some advanced models like SpanBERT and RoBERTa to assess their effectiveness in dialogue segmentation. We further experiment with optimization techniques to refine model performance. This analysis gives some insights for the future advancements in dialogue understanding and the development of more sophisticated conversation analysis systems.

You can access the project here

Nash Welfare and Stable Matchings

Supervisor:  Hadi Hosseini


Stable matching, originating from the College Admission problem by Gale and Shapley, encompasses applications ranging from kidney exchange to doctor-hospital and cadet branch matching. The Deferred Acceptance algorithm, foundational in this field, is criticized for consistently favoring one side of the market, prompting the development of fairer alternatives like Egalitarian, leximin, and Sex-Equal Stable matching. We analyzed the implications of Nash Welfare on one-to-one matching markets.



Detection of Breast Cancer based on Mammograms with Metadata Integration


Supervisor: Dr. C. Lee. Giles

    Collaborators:Amrit Puhan, Akshaya Jayant Patil

    Dept. of CSE, Pennsylvania State University

Conducted an empirical research project utilizing prominent CNN models (ResNet-50, ResNet-101, DenseNet-121) and Attention-based models (DeiT, DeiT-III, PatchConvnet) for predicting breast cancer from mammogram images, specifically employing the Kaggle RSNA Mammogram dataset. The innovative angle of our approach was the integration of image metadata during the deep learning models' training process. Results showcased that this metadata incorporation notably enhanced the AUC-ROC score across all models, while also improving generalization, pointing towards a viable avenue for advancing medical diagnostic AI.

Keywords: CNN, Attention-Based Models, Breast Cancer Prediction, Mammogram Image Analysis, Metadata Integration, Deep Learning.