arXiv subjects

Jialiang Mao

Publications and source records attributed to Jialiang Mao.

11 recordsLinked to original sources

arXiv2026

Multiscale Cochran-Mantel-Haenszel Scanning for Conditional Dependency

We propose a nonparametric approach to testing conditional independence and estimating conditional association, generalizing the Cochran-Mantel-Haenszel (CMH) test and odds-ratio estimator to continuous sample spaces. It leverages a multiscale scanning approach to decompose the sample space into a cascade of $2\times 2 \times T$ tables. Following the CMH test, we condition on the marginal order statistics, which are "almost ancillary" regarding conditional dependency. This strategy helps overcome a key challenge faced by other methods that discretize the sample space: we achieve consistency without requiring stratum sample sizes to grow to infinity, a constraint often difficult to satisfy in practice. Our method produces easy-to-compute test statistics with a known asymptotic null distribution under the conditional sampling model, scaling almost linearly with the sample size. Our simulation results demonstrate reliable Type I error control, even with small samples and high-dimensional conditioning, and competitive power compared to state-of-the-art tests. Finally, a case study on Uber ride-share data highlights the method's unique dual capability, inherited from the CMH, to both test and identify the nature of the inferred conditional association. By providing summary statistics that capture the strength and direction of local associations, our method offers practitioners a useful tool for learning conditional dependencies.

stat.ME↗

arXiv2025

TokenFLEX: Unified VLM Training for Flexible Visual Tokens Inference

Conventional Vision-Language Models(VLMs) typically utilize a fixed number of vision tokens, regardless of task complexity. This one-size-fits-all strategy introduces notable inefficiencies: using excessive tokens leads to unnecessary computational overhead in simpler tasks, whereas insufficient tokens compromise fine-grained visual comprehension in more complex contexts. To overcome these limitations, we present TokenFLEX, an innovative and adaptable vision-language framework that encodes images into a variable number of tokens for efficient integration with a Large Language Model (LLM). Our approach is underpinned by two pivotal innovations. Firstly, we present a novel training paradigm that enhances performance across varying numbers of vision tokens by stochastically modulating token counts during training. Secondly, we design a lightweight vision token projector incorporating an adaptive pooling layer and SwiGLU, allowing for flexible downsampling of vision tokens and adaptive selection of features tailored to specific token counts. Comprehensive experiments reveal that TokenFLEX consistently outperforms its fixed-token counterparts, achieving notable performance gains across various token counts enhancements of 1.6%, 1.0%, and 0.4% with 64, 144, and 256 tokens, respectively averaged over eight vision-language benchmarks. These results underscore TokenFLEX's remarkable flexibility while maintaining high-performance vision-language understanding.

cs.CV↗

arXiv2024

A tree-based model for addressing sparsity and taxa covariance in microbiome compositional count data

Microbiome compositional data are often high-dimensional, sparse, and exhibit pervasive cross-sample heterogeneity. Generative modeling is a popular approach to analyze such data, and effective generative models must accurately characterize these key features. While high-dimensionality and abundance of zeros have received much attention, existing models often lack flexibility in capturing complex cross-sample variability. This limitation can affect statistical efficiency and lead to misleading conclusions in tasks like differential abundance analysis, clustering, and network analysis. We introduce a generative model, the "logistic-tree normal" (LTN) model, which addresses this issue and effectively captures key characteristics of microbiome data, including abundance of zeros. LTN employs a tree-based decomposition to aggregate sparse taxa counts and uses a (multivariate) logistic-normal distribution at tree splits, allowing for flexible covariance adjustments among taxa as needed. The latent Gaussian structure of LTN enables the incorporation of multivariate analysis tools that enforce sparsity or low-rank covariance assumptions. As a versatile, fully generative model, LTN supports a wide range of applications and offers efficient Bayesian inference computational recipes through conjugate blocked Gibbs sampling with Pólya-Gamma augmentation. We demonstrate application of LTN in a compositional mixed-effects model for differential abundance analysis using numerical experiments and a reanalysis of the infant cohort in the DIABIMMUNE study. Our findings illustrate that LTN, by adequately accounting for cross-sample heterogeneity, appropriately generates the proportion of zeros without requiring an explicit zero-inflation component, confirming a recent viewpoint that "zero-inflation" in count-based sequencing data are often results of unaccounted cross-sample variation.

stat.ME↗

arXiv2023

Balancing Risk and Reward: An Automated Phased Release Strategy

Phased releases are a common strategy in the technology industry for gradually releasing new products or updates through a sequence of A/B tests in which the number of treated units gradually grows until full deployment or deprecation. Performing phased releases in a principled way requires selecting the proportion of units assigned to the new release in a way that balances the risk of an adverse effect with the need to iterate and learn from the experiment rapidly. In this paper, we formalize this problem and propose an algorithm that automatically determines the release percentage at each stage in the schedule, balancing the need to control risk while maximizing ramp-up speed. Our framework models the challenge as a constrained batched bandit problem that ensures that our pre-specified experimental budget is not depleted with high probability. Our proposed algorithm leverages an adaptive Bayesian approach in which the maximal number of units assigned to the treatment is determined by the posterior distribution, ensuring that the probability of depleting the remaining budget is low. Notably, our approach analytically solves the ramp sizes by inverting probability bounds, eliminating the need for challenging rare-event Monte Carlo simulation. It only requires computing means and variances of outcome subsets, making it highly efficient and parallelizable.

stat.ML↗

arXiv2023

Detecting Interference in A/B Testing with Increasing Allocation

In the past decade, the technology industry has adopted online randomized controlled experiments (a.k.a. A/B testing) to guide product development and make business decisions. In practice, A/B tests are often implemented with increasing treatment allocation: the new treatment is gradually released to an increasing number of units through a sequence of randomized experiments. In scenarios such as experimenting in a social network setting or in a bipartite online marketplace, interference among units may exist, which can harm the validity of simple inference procedures. In this work, we introduce a widely applicable procedure to test for interference in A/B testing with increasing allocation. Our procedure can be implemented on top of an existing A/B testing platform with a separate flow and does not require a priori a specific interference mechanism. In particular, we introduce two permutation tests that are valid under different assumptions. Firstly, we introduce a general statistical test for interference requiring no additional assumption. Secondly, we introduce a testing procedure that is valid under a time fixed effect assumption. The testing procedure is of very low computational complexity, it is powerful, and it formalizes a heuristic algorithm implemented already in industry. We demonstrate the performance of the proposed testing procedure through simulations on synthetic data. Finally, we discuss one application at LinkedIn, where a screening step is implemented to detect potential interference in all their marketplace experiments with the proposed methods in the paper.

stat.ME↗

arXiv2022

Causal Estimation of Position Bias in Recommender Systems Using Marketplace Instruments

Information retrieval systems, such as online marketplaces, news feeds, and search engines, are ubiquitous in today's digital society. They facilitate information discovery by ranking retrieved items on predicted relevance, i.e. likelihood of interaction (click, share) between users and items. Typically modeled using past interactions, such rankings have a major drawback: interaction depends on the attention items receive. A highly-relevant item placed outside a user's attention could receive little interaction. This discrepancy between observed interaction and true relevance is termed the position bias. Position bias degrades relevance estimation and when it compounds over time, it can silo users into false relevant items, causing marketplace inefficiencies. Position bias may be identified with randomized experiments, but such an approach can be prohibitive in cost and feasibility. Past research has also suggested propensity score methods, which do not adequately address unobserved confounding; and regression discontinuity designs, which have poor external validity. In this work, we address these concerns by leveraging the abundance of A/B tests in ranking evaluations as instrumental variables. Historical A/B tests allow us to access exogenous variation in rankings without manually introducing them, harming user experience and platform revenue. We demonstrate our methodology in two distinct applications at LinkedIn - feed ads and the People-You-May-Know (PYMK) recommender. The marketplaces comprise users and campaigns on the ads side, and invite senders and recipients on PYMK. By leveraging prior experimentation, we obtain quasi-experimental variation in item rankings that is orthogonal to user relevance. Our method provides robust position effect estimates that handle unobserved confounding well, greater generalizability, and easily extends to other information retrieval systems.

econ.EM↗

arXiv2021

Quantifying the Value of Iterative Experimentation

Over the past decade, most technology companies and a growing number of conventional firms have adopted online experimentation (or A/B testing) into their product development process. Initially, A/B testing was deployed as a static procedure in which an experiment was conducted by randomly splitting half of the users to see the control-the standard offering-and the other half the treatment-the new version. The results were then used to augment decision-making around which version to release widely. More recently, as experimentation has matured, firms have developed a more dynamic approach to experimentation in which a new version (the treatment) is gradually released to a growing number of units through a sequence of randomized experiments, known as iterations. In this paper, we develop a theoretical framework to quantify the value brought on by such dynamic or iterative experimentation. We apply our framework to seven months of LinkedIn experiments and show that iterative experimentation led to an additional 20% improvement in one of the firm's primary metrics.

stat.AP↗

arXiv2020

Trustworthy Online Marketplace Experimentation with Budget-split Design

Online experimentation, also known as A/B testing, is the gold standard for measuring product impacts and making business decisions in the tech industry. The validity and utility of experiments, however, hinge on unbiasedness and sufficient power. In two-sided online marketplaces, both requirements are called into question. The Bernoulli randomized experiments are biased because treatment units interfere with control units through market competition and violate the "stable unit treatment value assumption"(SUTVA). The experimental power on at least one side of the market is often insufficient because of disparate sample sizes on the two sides. Despite the important of online marketplaces to the online economy and the crucial role experimentation plays in product improvement, there lacks an effective and practical solution to the bias and low power problems in marketplace experimentation. Our paper fills this gap by proposing an experimental design that is unbiased in any marketplace where buyers have a defined budget, which could be finite or infinite. We show that it is more powerful than all other unbiased designs in literature. We then provide generalizable system architecture for deploying this design to online marketplaces. Finally, we confirm our findings with empirical performance from experiments run in two real-world online marketplaces.

stat.AP↗

arXiv2020

Dirichlet-tree multinomial mixtures for clustering microbiome compositions

Studying the human microbiome has gained substantial interest in recent years, and a common task in the analysis of these data is to cluster microbiome compositions into subtypes. This subdivision of samples into subgroups serves as an intermediary step in achieving personalized diagnosis and treatment. In applying existing clustering methods to modern microbiome studies including the American Gut Project (AGP) data, we found that this seemingly standard task, however, is very challenging in the microbiome composition context due to several key features of such data. Standard distance-based clustering algorithms generally do not produce reliable results as they do not take into account the heterogeneity of the cross-sample variability among the bacterial taxa, while existing model-based approaches do not allow sufficient flexibility for the identification of complex within-cluster variation from cross-cluster variation. Direct applications of such methods generally lead to overly dispersed clusters in the AGP data and such phenomenon is common for other microbiome data. To overcome these challenges, we introduce Dirichlet-tree multinomial mixtures (DTMM) as a Bayesian generative model for clustering amplicon sequencing data in microbiome studies. DTMM models the microbiome population with a mixture of Dirichlet-tree kernels that utilizes the phylogenetic tree to offer a more flexible covariance structure in characterizing within-cluster variation, and it provides a means for identifying a subset of signature taxa that distinguish the clusters. We perform extensive simulation studies to evaluate the performance of DTMM and compare it to state-of-the-art model-based and distance-based clustering methods in the microbiome context. Finally, we report a case study on the fecal data from the AGP to identify compositional clusters among individuals with inflammatory bowel disease and diabetes.

stat.ME↗

arXiv2019

Bayesian graphical compositional regression for microbiome data

An important task in microbiome studies is to test the existence of and give characterization to differences in the microbiome composition across groups of samples. Important challenges of this problem include the large within-group heterogeneities among samples and the existence of potential confounding variables that, when ignored, increase the chance of false discoveries and reduce the power for identifying true differences. We propose a probabilistic framework to overcome these issues by combining three ideas: (i) a phylogenetic tree-based decomposition of the cross-group comparison problem into a series of local tests, (ii) a graphical model that links the local tests to allow information sharing across taxa, and (iii) a Bayesian testing strategy that incorporates covariates and integrates out the within-group variation, avoiding potentially unstable point estimates. We derive an efficient inference algorithm based on numerical integration and junction-tree message passing, conduct extensive simulation studies to investigate the performance of our approach, and compare it to state-of-the-art methods in a number of representative settings. We then apply our method to the American Gut data to analyze the association of dietary habits and human's gut microbiome composition in the presence of covariates, and illustrate the importance of incorporating covariates in microbiome cross-group comparison.

stat.ME↗

arXiv2017

Fisher exact scanning for dependency

We introduce a method---called Fisher exact scanning (FES)---for testing and identifying variable dependency that generalizes Fisher's exact test on $2\times 2$ contingency tables to $R\times C$ contingency tables and continuous sample spaces. FES proceeds through scanning over the sample space using windows in the form of $2\times 2$ tables of various sizes, and on each window completing a Fisher's exact test. Based on a factorization of Fisher's multivariate hypergeometric (MHG) likelihood into the product of the univariate hypergeometric likelihoods, we show that there exists a coarse-to-fine, sequential generative representation for the MHG model in the form of a Bayesian network, which in turn implies the mutual independence (up to deviation due to discreteness) among the Fisher's exact tests completed under FES. This allows an exact characterization of the joint null distribution of the $p$-values and gives rise to an effective inference recipe through simple multiple testing procedures such as Šidák and Bonferroni corrections, eliminating the need for resampling. In addition, FES can characterize dependency through reporting significant windows after multiple testing control. The computational complexity of FES is approximately linear in the sample size, which along with the avoidance of resampling makes it ideal for analyzing massive data sets. We use extensive numerical studies to illustrate the work of FES and compare it to several state-of-the-art methods for testing dependency in both statistical and computational performance. Finally, we apply FES to analyzing a microbiome data set and further investigate its relationship with other popular dependency metrics in that context.

stat.ME↗