arXiv subjects

Yuhao Du

Publications and source records attributed to Yuhao Du.

26 records · Page 2Linked to original sources

arXiv2023

Comprehensive Visual Question Answering on Point Clouds through Compositional Scene Manipulation

Visual Question Answering on 3D Point Cloud (VQA-3D) is an emerging yet challenging field that aims at answering various types of textual questions given an entire point cloud scene. To tackle this problem, we propose the CLEVR3D, a large-scale VQA-3D dataset consisting of 171K questions from 8,771 3D scenes. Specifically, we develop a question engine leveraging 3D scene graph structures to generate diverse reasoning questions, covering the questions of objects' attributes (i.e., size, color, and material) and their spatial relationships. Through such a manner, we initially generated 44K questions from 1,333 real-world scenes. Moreover, a more challenging setup is proposed to remove the confounding bias and adjust the context from a common-sense layout. Such a setup requires the network to achieve comprehensive visual understanding when the 3D scene is different from the general co-occurrence context (e.g., chairs always exist with tables). To this end, we further introduce the compositional scene manipulation strategy and generate 127K questions from 7,438 augmented 3D scenes, which can improve VQA-3D models for real-world comprehension. Built upon the proposed dataset, we baseline several VQA-3D models, where experimental results verify that the CLEVR3D can significantly boost other 3D scene understanding tasks. Our code and dataset will be made publicly available at https://github.com/yanx27/CLEVR3D.

cs.CV↗

arXiv2022

Multi-Modal Masked Autoencoders for Medical Vision-and-Language Pre-Training

Medical vision-and-language pre-training provides a feasible solution to extract effective vision-and-language representations from medical images and texts. However, few studies have been dedicated to this field to facilitate medical vision-and-language understanding. In this paper, we propose a self-supervised learning paradigm with multi-modal masked autoencoders (M$^3$AE), which learn cross-modal domain knowledge by reconstructing missing pixels and tokens from randomly masked images and texts. There are three key designs to make this simple approach work. First, considering the different information densities of vision and language, we adopt different masking ratios for the input image and text, where a considerably larger masking ratio is used for images. Second, we use visual and textual features from different layers to perform the reconstruction to deal with different levels of abstraction in visual and language. Third, we develop different designs for vision and language decoders (i.e., a Transformer for vision and a multi-layer perceptron for language). To perform a comprehensive evaluation and facilitate further research, we construct a medical vision-and-language benchmark including three tasks. Experimental results demonstrate the effectiveness of our approach, where state-of-the-art results are achieved on all downstream tasks. Besides, we conduct further analysis to better verify the effectiveness of different components of our approach and various settings of pre-training. The source code is available at~\url{https://github.com/zhjohnchan/M3AE}.

cs.CV↗

arXiv2022

A Data-Driven Simulation of the New York State Foster Care System

We introduce an analytic pipeline to model and simulate youth trajectories through the New York state foster care system. Our goal in doing so is to forecast how proposed interventions may impact the foster care system's ability to achieve it's stated goals \emph{before these interventions are actually implemented and impact the lives of thousands of youth}. Here, we focus on two specific stated goals of the system: racial equity, and, as codified most recently by the 2018 Family First Prevention Services Act (FFPSA), a focus on keeping all youth out of foster care. We also focus on one specific potential intervention -- a predictive model, proposed in prior work and implemented elsewhere in the U.S., which aims to determine whether or not a youth is in need of care. We use our method to explore how the implementation of this predictive model in New York would impact racial equity and the number of youth in care. While our findings, as in any simulation model, ultimately rely on modeling assumptions, we find evidence that the model would not necessarily achieve either goal. Primarily, then, we aim to further promote the use of data-driven simulation to help understand the ramifications of algorithmic interventions in public systems.

cs.CY↗

arXiv2021

A Computational Social Science Approach to Understanding Predictors of Chafee Service Receipt

The John H. Chafee Foster Care Program for Successful Transition to Adulthood (CFCIP) allocates funding to provide services to youth who are likely to age out of foster care. These services, covering everything from mentoring to financial aid, are expected to be distributed in ways that prepare youth for life after care. However, surprisingly little is known about which youth receive which services. The present work makes use of the National Youth in Transition Database (NYTD), a large-scale administrative dataset that tracks services allocated to youth that use CFCIP funds. Specifically, we conduct a forensic social science analysis of the NYTD data. To do so, we first use computational methods to help us uncover the most important factors associated with service receipt. Doing so helps us to identify three major factors-youth age, youth time in care, and the state in which a youth is in care-that are most heavily associated with service receive. We then conduct an analysis that links existing theory to these factors, expanding our understanding of how services are allocated and paving the way to future work to understand why such associations exist.

cs.SI↗

arXiv2021

Insidious Nonetheless: How Small Effects and Hierarchical Norms Create and Maintain Gender Disparities in Organizations

The term glass ceiling is applied to the well-established phenomenon in which women and people of color are consistently blocked from reaching the upper-most levels of the corporate hierarchy. Focusing on gender, we present an agent-based model that explores how empirically established mechanisms of interpersonal discrimination coevolve with social norms at both the organizational (meso) and societal (macro) levels to produce this glass ceiling effect for women. Our model extends our understanding of how the glass ceiling arises, and why it can be resistant to change. We do so by synthesizing existing psychological and structural theories of discrimination into a mathematical model that quantifies explicitly how complex organizational systems can produce and maintain inequality. We discuss implications of our findings for both intervention and future empirical analyses, and provide open-source code for those wishing to adapt or extend our work.

cs.SI↗

arXiv2021

Incentives in Two-sided Matching Markets with Prediction-enhanced Preference-formation

Two-sided matching markets have long existed to pair agents in the absence of regulated exchanges. A common example is school choice, where a matching mechanism uses student and school preferences to assign students to schools. In such settings, forming preferences is both difficult and critical. Prior work has suggested various prediction mechanisms that help agents make decisions about their preferences. Although often deployed together, these matching and prediction mechanisms are almost always analyzed separately. The present work shows that at the intersection of the two lies a previously unexplored type of strategic behavior: agents returning to the market (e.g., schools) can attack future predictions by interacting short-term non-optimally with their matches. Here, we first introduce this type of strategic behavior, which we call an `adversarial interaction attack'. Next, we construct a formal economic model that captures the feedback loop between prediction mechanisms designed to assist agents and the matching mechanism used to pair them. This economic model allows us to analyze adversarial interaction attacks. Finally, using school choice as an example, we build a simulation to show that, as the trust in and accuracy of predictions increases, schools gain progressively more by initiating an adversarial interaction attack. We also show that this attack increases inequality in the student population.

cs.LG↗

arXiv2020

MDR Cluster-Debias: A Nonlinear WordEmbedding Debiasing Pipeline

Existing methods for debiasing word embeddings often do so only superficially, in that words that are stereotypically associated with, e.g., a particular gender in the original embedding space can still be clustered together in the debiased space. However, there has yet to be a study that explores why this residual clustering exists, and how it might be addressed. The present work fills this gap. We identify two potential reasons for which residual bias exists and develop a new pipeline, MDR Cluster-Debias, to mitigate this bias. We explore the strengths and weaknesses of our method, finding that it significantly outperforms other existing debiasing approaches on a variety of upstream bias tests but achieves limited improvement on decreasing gender bias in a downstream task. This indicates that word embeddings encode gender bias in still other ways, not necessarily captured by upstream tests.

cs.CL↗

arXiv2018

Improved Algorithms for Fully Dynamic Maximal Independent Set

Maintaining maximal independent set in dynamic graph is a fundamental open problem in graph theory and the first sublinear time deterministic algorithm was came up by Assadi, Onak, Schieber and Solomon(STOC'18), which achieves $O(m^{3/4})$ amortized update time. We have two main contributions in this paper. We present a new simple deterministic algorithm with $O(m^{2/3}\sqrt{\log m})$ amortized update time, which improves the previous best result. And we also present the first randomized algorithm with expected $O(\sqrt{m}\log^{1.5}m)$ amortized time against an oblivious adversary.

cs.DS↗