arXiv subjects

Yang Ni

Publications and source records attributed to Yang Ni.

At least 37 records · Page 2Linked to original sources

arXiv2026

Multivariable Bidirectional Mendelian Randomization via Bayesian Directed Cyclic Graphical Models with Correlated Errors

Mendelian randomization (MR) is a pivotal tool in genetics, genomics, and epidemiology, leveraging genetic variants as instrumental variables to infer causal relationships between exposures and outcomes. Traditional MR methods, while powerful, often rely on stringent assumptions such as the absence of feedback loops, which are frequently violated in complex biological networks. In addition, many popular MR approaches focus on only two variables (i.e., one exposure and one outcome), whereas our motivating applications of gene regulatory networks have many variables. In this article, we introduce a novel Bayesian framework for multivariable MR that concurrently addresses unmeasured confounding and feedback loops. Central to our approach is a sparse conditional cyclic graphical model with a sparse error variance-covariance matrix. Two structural priors are employed to enable the modeling and inference of causal relationships as well as latent confounding structures. Our method is designed to operate effectively with summary-level data, facilitating its application in contexts where individual-level data are inaccessible, e.g., due to privacy concerns. It can also account for horizontal pleiotropy, under which we establish the sufficient identifiability conditions. Through extensive simulations and applications to the GTEx and OneK1K data, we demonstrate the superior performance of our approach in recovering biologically plausible causal relationships in the presence of possible feedback loops and unmeasured confounding. Using posterior samples, we further quantify uncertainty in inferred network motifs by computing their posterior probabilities. The R package MR.RGM that implements the proposed method is available on CRAN (https://cran.r-project.org/package=MR.RGM).

stat.ME↗

arXiv2026

Bayesian Multiple Multivariate Density-Density Regression

We propose the first approach for multiple multivariate density-density regression (MDDR), making it possible to consider the regression of a multivariate density-valued response on multiple multivariate density-valued predictors. The core idea is to define a fitted distribution using a sliced Wasserstein barycenter (SWB) of push-forwards of the predictors and to quantify deviations from the observed response using the sliced Wasserstein (SW) distance. Regression functions, which map predictors' supports to the response support, and barycenter weights are inferred within a generalized Bayes framework, enabling principled uncertainty quantification without requiring a fully specified likelihood. The inference process can be seen as an instance of an inverse SWB problem. We establish theoretical guarantees, including the stability of the SWB under perturbations of marginals and barycenter weights, sample complexity of the generalized likelihood, and posterior consistency. For practical inference, we introduce a differentiable approximation of the SWB and a smooth reparameterization to handle the simplex constraint on barycenter weights, allowing efficient gradient-based MCMC sampling. We demonstrate MDDR in an application to inference for population-scale single-cell data. Posterior analysis under the MDDR model in this example includes inference on communication between multiple source/sender cell types and a target/receiver cell type. The proposed approach provides accurate fits, reliable predictions, and interpretable posterior estimates of barycenter weights, which can be used to construct sparse cell-cell communication networks.

stat.ME↗

arXiv2025

"Even GPT Can Reject Me": Conceptualizing Abrupt Refusal Secondary Harm (ARSH) and Reimagining Psychological AI Safety with Compassionate Completion Standard (CCS)

Large Language Models (LLMs) and AI chatbots are increasingly used for emotional and mental health support due to their low cost, immediacy, and accessibility. However, when safety guardrails are triggered, conversations may be abruptly terminated, introducing a distinct form of emotional disruption that can exacerbate distress and elevate risk among already vulnerable users. As this phenomenon gains attention, this viewpoint introduces Abrupt Refusal Secondary Harm (ARSH) as a conceptual framework to describe the psychological impacts of sudden conversational discontinuation caused by AI safety protocols. Drawing on counseling psychology and communication science as conceptual heuristics, we argue that abrupt refusals can rupture perceived relational continuity, evoke feelings of rejection or shame, and discourage future help seeking. To mitigate these risks, we propose a design hypothesis, the Compassionate Completion Standard (CCS), a refusal protocol grounded in Human Centered Design (HCD) that maintains safety constraints while preserving relational coherence. CCS emphasizes empathetic acknowledgment, transparent boundary articulation, graded conversational transition, and guided redirection, replacing abrupt disengagement with psychologically attuned closure. By integrating awareness of ARSH into AI safety design, developers and policymakers can reduce preventable iatrogenic harm and advance a more psychologically informed approach to AI governance. Rather than presenting incremental empirical findings, this viewpoint contributes a timely conceptual framework, articulates a testable design hypothesis, and outlines a coordinated research agenda for improving psychological safety in human AI interaction.

cs.CY↗

arXiv2025

A Dependent Feature Allocation Model Based on Random Fields

We introduce a flexible framework for modeling dependent feature allocations. Our approach addresses limitations in traditional nonparametric methods by directly modeling the logit-probability surface of the feature paintbox, enabling the explicit incorporation of covariates and complex but tractable dependence structures. The core of our model is a Gaussian Markov Random Field (GMRF), which we use to robustly decompose the latent field, separating a structural component based on the baseline covariates from intrinsic, unstructured heterogeneity. This structure is not a rigid grid but a sparse k-nearest neighbors graph derived from the latent geometry in the data, ensuring high-dimensional tractability. We extend this framework to a dynamic spatio-temporal process, allowing item effects to evolve via an Ornstein-Uhlenbeck process. Feature correlations are captured using a low-rank factorization of their joint prior. We demonstrate our model's utility by applying it to a polypharmacy dataset, successfully inferring latent health conditions from patient drug profiles.

stat.ME↗

arXiv2025

Spatially Varying Gene Regulatory Networks via Bayesian Nonparametric Covariate-Dependent Directed Cyclic Graphical Models

Spatial transcriptomics technologies enable the measurement of gene expression with spatial context, providing opportunities to understand how gene regulatory networks vary across tissue regions. However, existing graphical models focus primarily on undirected graphs or directed acyclic graphs, limiting their ability to capture feedback loops that are prevalent in gene regulation. Moreover, ensuring the so-called stability condition of cyclic graphs, while allowing graph structures to vary continuously with spatial covariates, presents significant statistical and computational challenges. We propose BNP-DCGx, a Bayesian nonparametric approach for learning spatially varying gene regulatory networks via covariate-dependent directed cyclic graphical models. Our method introduces a covariate-dependent random partition as an intermediary layer in a hierarchical model, which discretizes the covariate space into clusters with cluster-specific stable directed cyclic graphs. Through partition averaging, we obtain smoothly varying graph structures over space while maintaining theoretical guarantees of stability. We develop an efficient parallel tempered Markov chain Monte Carlo algorithm for posterior inference and demonstrate through simulations that our method accurately recovers both piecewise constant and continuously varying graph structures. Application to spatial transcriptomics data from human dorsolateral prefrontal cortex reveals spatially varying regulatory networks with feedback loops, identifies potential cell subtypes within established cell types based on distinct regulatory mechanisms, and provides new insights into spatial organization of gene regulation in brain tissue.

stat.ME↗

arXiv2025

Cauchy-Schwarz Fairness Regularizer

Group fairness in machine learning is often enforced by adding a regularizer that reduces the dependence between model predictions and sensitive attributes. However, existing regularizers are built on heterogeneous distance measures and design choices, which makes their behavior hard to reason about and their performance inconsistent across tasks. This raises a basic question: what properties make a good fairness regularizer? We address this question by first organizing existing in-process methods into three families: (i) matching prediction statistics across sensitive groups, (ii) aligning latent representations, and (iii) directly minimizing dependence between predictions and sensitive attributes. Through this lens, we identify desirable properties of the underlying distance measure, including tight generalization bounds, robustness to scale differences, and the ability to handle arbitrary prediction distributions. Motivated by these properties, we propose a Cauchy-Schwarz (CS) fairness regularizer that penalizes the empirical CS divergence between prediction distributions conditioned on sensitive groups. Under a Gaussian comparison, we show that CS divergence yields a tighter bound than Kullback-Leibler divergence, Maximum Mean Discrepancy, and the mean disparity used in Demographic Parity, and we discuss how these advantages translate to a distribution-free, kernel-based estimator that naturally extends to multiple sensitive attributes. Extensive experiments on four tabular benchmarks and one image dataset demonstrate that the proposed CS regularizer consistently improves Demographic Parity and Equal Opportunity metrics while maintaining competitive accuracy, and achieves a more stable utility-fairness trade-off across hyperparameter settings compared to prior regularizers.

cs.LG↗

arXiv2025

Recover-to-Forget: Gradient Reconstruction from LoRA for Efficient LLM Unlearning

Unlearning in large foundation models (e.g., LLMs) is essential for enabling dynamic knowledge updates, enforcing data deletion rights, and correcting model behavior. However, existing unlearning methods often require full-model fine-tuning or access to the original training data, which limits their scalability and practicality. In this work, we introduce Recover-to-Forget (R2F), a novel framework for efficient unlearning in LLMs based on reconstructing full-model gradient directions from low-rank LoRA adapter updates. Rather than performing backpropagation through the full model, we compute gradients with respect to LoRA parameters using multiple paraphrased prompts and train a gradient decoder to approximate the corresponding full-model gradients. To ensure applicability to larger or black-box models, the decoder is trained on a proxy model and transferred to target models. We provide a theoretical analysis of cross-model generalization and demonstrate that our method achieves effective unlearning while preserving general model performance. Experimental results demonstrate that R2F offers a scalable and lightweight alternative for unlearning in pretrained LLMs without requiring full retraining or access to internal parameters.

cs.LG↗

arXiv2025

LUNE: Efficient LLM Unlearning via LoRA Fine-Tuning with Negative Examples

Large language models (LLMs) possess vast knowledge acquired from extensive training corpora, but they often cannot remove specific pieces of information when needed, which makes it hard to handle privacy, bias mitigation, and knowledge correction. Traditional model unlearning approaches require computationally expensive fine-tuning or direct weight editing, making them impractical for real-world deployment. In this work, we introduce LoRA-based Unlearning with Negative Examples (LUNE), a lightweight framework that performs negative-only unlearning by updating only low-rank adapters while freezing the backbone, thereby localizing edits and avoiding disruptive global changes. Leveraging Low-Rank Adaptation (LoRA), LUNE targets intermediate representations to suppress (or replace) requested knowledge with an order-of-magnitude lower compute and memory than full fine-tuning or direct weight editing. Extensive experiments on multiple factual unlearning tasks show that LUNE: (I) achieves effectiveness comparable to full fine-tuning and memory-editing methods, and (II) reduces computational cost by about an order of magnitude.

cs.LG↗

arXiv2025

Mitigating Bias in Graph Hyperdimensional Computing

Graph hyperdimensional computing (HDC) has emerged as a promising paradigm for cognitive tasks, emulating brain-like computation with high-dimensional vectors known as hypervectors. While HDC offers robustness and efficiency on graph-structured data, its fairness implications remain largely unexplored. In this paper, we study fairness in graph HDC, where biases in data representation and decision rules can lead to unequal treatment of different groups. We show how hypervector encoding and similarity-based classification can propagate or even amplify such biases, and we propose a fairness-aware training framework, FairGHDC, to mitigate them. FairGHDC introduces a bias correction term, derived from a gap-based demographic-parity regularizer, and converts it into a scalar fairness factor that scales the update of the class hypervector for the ground-truth label. This enables debiasing directly in the hypervector space without modifying the graph encoder or requiring backpropagation. Experimental results on six benchmark datasets demonstrate that FairGHDC substantially reduces demographic-parity and equal-opportunity gaps while maintaining accuracy comparable to standard GNNs and fairness-aware GNNs. At the same time, FairGHDC preserves the computational advantages of HDC, achieving up to about one order of magnitude ($\approx 10\times$) speedup in training time on GPU compared to GNN and fairness-aware GNN baselines.

cs.LG↗

arXiv2025

Exploring ChatGPT's Capabilities, Stability, Potential and Risks in Conducting Psychological Counseling through Simulations in School Counseling

This study explores ChatGPT's capabilities, stability, and risks in simulating psychological counseling sessions in a school counseling context. Using scripted role-plays between a human counselor and an AI client, we examine how a large language model performs core counseling skills such as empathy, reflection, summarizing, and asking open-ended questions, as well as its ability to maintain therapeutic communication over time. We focus on how consistently ChatGPT can behave like a "virtual client" for school counselors in training, and how its responses might support or disrupt counselor skill development, supervision, and practice. At the same time, we analyze potential risks, including inaccurate or unsafe suggestions, over-compliance with counselor prompts, and the illusion of a competent therapist where no real professional judgment exists. The findings suggest that ChatGPT can serve as a low-cost, always-available training tool for practicing counseling techniques and interviewing skills in education and mental health settings, but it should not be viewed as a replacement for a human therapist or school counselor. We propose practical guidelines for educators, supervisors, and researchers who wish to use ChatGPT or similar LLM-based conversational agents in counseling training, highlighting how to leverage its potential while managing ethical, pedagogical, and psychological risks.

cs.HC↗

arXiv2025

QUILL: An Algorithm-Architecture Co-Design for Cache-Local Deformable Attention

Deformable transformers deliver state-of-the-art detection but map poorly to hardware due to irregular memory access and low arithmetic intensity. We introduce QUILL, a schedule-aware accelerator that turns deformable attention into cache-friendly, single-pass work. At its core, Distance-based Out-of-Order Querying (DOOQ) orders queries by spatial proximity; the look-ahead drives a region prefetch into an alternate buffer--forming a schedule-aware prefetch loop that overlaps memory and compute. A fused MSDeformAttn engine executes interpolation, Softmax, aggregation, and the final projection (W''m) in one pass without spilling intermediates, while small tensors are kept on-chip and surrounding dense layers run on integrated GEMMs. Implemented as RTL and evaluated end-to-end, QUILL achieves up to 7.29x higher throughput and 47.3x better energy efficiency than an RTX 4090, and exceeds prior accelerators by 3.26-9.82x in throughput and 2.01-6.07x in energy efficiency. With mixed-precision quantization, accuracy tracks FP32 within <=0.9 AP across Deformable and Sparse DETR variants. By converting sparsity into locality--and locality into utilization--QUILL delivers consistent, end-to-end speedups.

cs.AR↗

arXiv2025

RETuning: Upgrading Inference-Time Scaling for Stock Movement Prediction with Large Language Models

Recently, large language models (LLMs) have demonstrated outstanding reasoning capabilities on mathematical and coding tasks. However, their application to financial tasks-especially the most fundamental task of stock movement prediction-remains underexplored. We study a three-class classification problem (up, hold, down) and, by analyzing existing reasoning responses, observe that: (1) LLMs follow analysts' opinions rather than exhibit a systematic, independent analytical logic (CoTs). (2) LLMs list summaries from different sources without weighing adversarial evidence, yet such counterevidence is crucial for reliable prediction. It shows that the model does not make good use of its reasoning ability to complete the task. To address this, we propose Reflective Evidence Tuning (RETuning), a cold-start method prior to reinforcement learning, to enhance prediction ability. While generating CoT, RETuning encourages dynamically constructing an analytical framework from diverse information sources, organizing and scoring evidence for price up or down based on that framework-rather than on contextual viewpoints-and finally reflecting to derive the prediction. This approach maximally aligns the model with its learned analytical framework, ensuring independent logical reasoning and reducing undue influence from context. We also build a large-scale dataset spanning all of 2024 for 5,123 A-share stocks, with long contexts (32K tokens) and over 200K samples. In addition to price and news, it incorporates analysts' opinions, quantitative reports, fundamental data, macroeconomic indicators, and similar stocks. Experiments show that RETuning successfully unlocks the model's reasoning ability in the financial domain. Inference-time scaling still works even after 6 months or on out-of-distribution stocks, since the models gain valuable insights about stock movement prediction.

cs.CL↗

arXiv2025

RIGEL: Feedback-regulated cloud-scale star formation efficiency in a simulated dwarf galaxy merger

Major mergers of galaxies are likely to trigger bursty star formation activities. The accumulation of dense gas and the boost of star formation efficiency (SFE) are considered to be the two main drivers of starbursts. However, it remains unclear how each process operates on the scale of individual star-forming clouds. Here, we present a high-resolution (2 Msun) RHD simulation of a gas-rich dwarf galaxy merger using the RIGEL model to investigate how mergers affect the properties of the structure of dense star-forming gas and the cloud-scale SFE. We tracked the evolution of sub-virial dense clouds in the simulation by mapping them across successive snapshots taken at intervals of 0.2 Myr. We find that the merger triggers a 130 fold increase in the SFR and shortens the galaxy-wide gas-depletion time by two orders of magnitude compared to those in two isolated galaxies. However, the depletion time of individual clouds and their lifetime distribution remained unchanged over the simulation period. The cloud life cycles and cloud-scale SFE are determined by the local stellar feedback rather than such environmental factors as tidal fields regardless of the merger process, and the integrated SFE ($ε_{\rm int}$) of clouds in complex environments remains well-described by an $ε_{\rm int}-Σ_{\rm tot}$ relation found in idealized isolated-cloud experiments. During the peak of the starburst, the media SFE was lower by only 0.17-0.33 dex compared to the value when the galaxies were not interacting. The merger boosts the SFR through the accumulation and compression of dense gas fueling star formation. Strong tidal torques assemble $>10^5$ Msun clouds, which seed massive star clusters. The average separation between star-forming clouds decreases during the merger, which in turn decreases the cloud--cluster spatial de-correlation from >1 kpc to 0.1 kpc depicted in tuning fork diagrams.

astro-ph.GA↗

arXiv2025

Radiation Hydrodynamics of Self-gravitating Protoplanetary Disks: I. Direct Formation of Gas Giants via Disk Fragmentation

Gravitational instability (GI) has long been considered a viable pathway for giant planet formation in protoplanetary disks (PPDs), especially at wide orbital separations or around low-mass stars where core accretion faces significant challenges. However, a primary drawback is that disk fragmentation from GI was generally found to produce over-massive clumps, typically in the mass range of brown dwarfs, although most numerical studies adopted simplified cooling prescriptions or with limited numerical resolution. We conduct a suite of global three-dimensional radiation hydrodynamics (RHD) simulations of self-gravitating PPDs using the meshless finite-mass (MFM) method. By implementing radiation transport via the M1 closure and systematically varying disk mass and opacity, we show that increasing disk mass and lowering opacity promote fragmentation by enhancing radiative cooling. Non-fragmenting disks settle into a gravito-turbulent state with low-order spiral structures and effective angular momentum transport characterized by $α\sim β_\mathrm{cool}^{-1}$. In fragmenting disks, a subset of gravitationally bound clumps survives as long-lived fragments. Their initial masses form a consistent distribution around $Σ\cdotλ_\mathrm{T} \cdot 2 (c_s/Ω_\mathrm{K})$ (with $λ_T$ the Toomre wavelength), corresponding to $\sim 0.3 - 10\,M_\mathrm{J}$ in our simulations, consistent with being gas giants. These results demonstrate that GI can produce planet-mass fragments under more realistic conditions, reinforcing it as a viable gas giant formation pathway and motivating further studies of fragment evolution and observational signatures.

astro-ph.EP↗

arXiv2025

Global-Local Dirichlet Processes for Identifying Pan-Cancer Subpopulations Using Both Shared and Cancer-Specific Data

We consider the problem of clustering grouped data for which the observations may include group-specific variables in addition to the variables that are shared across groups. This type of data is common in cancer genomics where the molecular information is usually accompanied by cancer-specific clinical information. Existing grouped clustering methods only consider the shared variables, thereby ignoring valuable information from the cancer-specific variables. To allow for these cancer-specific variables to aid in the clustering, we propose a novel Bayesian nonparametric approach, termed global-local (GLocal) Dirichlet process, that models the ``global-local'' structure of the observations across groups. We characterize the GLocal Dirichlet process using the stick-breaking representation and the representation as a limit of a finite mixture model, which leads to an efficient posterior inference algorithm. We illustrate our model with extensive simulations and a real pan-gastrointestinal cancer dataset. The cancer-specific clinical variables included carcinoembryonic antigen level, patients' body mass index, and the number of cigarettes smoked per day. These important clinical variables refine the clusters of gene expression data and allow us to identify finer sub-clusters, which is not possible in their absence. This refinement aids in the better understanding of tumor progression and heterogeneity. Moreover, our proposed method is applicable beyond the field of cancer genomics to a general grouped clustering framework in the presence of group-specific idiosyncratic variables.

stat.ME↗

arXiv2025

Bayesian Multivariate Density-Density Regression

We introduce a novel and scalable Bayesian framework for multivariate-density-density regression (DDR), designed to model relationships between multivariate distributions. Our approach addresses the critical issue of distributions residing in spaces of differing dimensions. We utilize a generalized Bayes framework, circumventing the need for a fully specified likelihood by employing the sliced Wasserstein distance to measure the discrepancy between fitted and observed distributions. This choice not only handles high-dimensional data and varying sample sizes efficiently but also facilitates a Metropolis-adjusted Langevin algorithm (MALA) for posterior inference. Furthermore, we establish the posterior consistency of our generalized Bayesian approach, ensuring that the posterior distribution concentrates around the true parameters as the sample size increases. Through simulations and application to a population-scale single-cell dataset, we show that Bayesian DDR provides robust fits, superior predictive performance compared to traditional methods, and valuable insights into complex biological interactions.

stat.ME↗

arXiv2025

Consistent DAG selection for Bayesian causal discovery under general error distributions

We consider the problem of learning the underlying causal structure among a set of variables, which are assumed to follow a Bayesian network or, more specifically, a linear recursive structural equation model (SEM) with the associated errors being independent and allowed to be non-Gaussian. A Bayesian hierarchical model is proposed to identify the true data-generating directed acyclic graph (DAG) structure where the nodes and edges represent the variables and the direct causal effects, respectively. Moreover, incorporating the information of non-Gaussian errors, we characterize the distribution equivalence class of the true DAG, which specifies the best possible extent to which the DAG can be identified based on purely observational data. Furthermore, under the consideration that the errors are distributed as some scale mixture of Gaussian, where the mixing distribution is unspecified, and mild distributional assumptions, we establish that by employing a non-standard DAG prior, the posterior probability of the distribution equivalence class of the true DAG converges to unity as the sample size grows. This shows that the proposed method achieves the posterior DAG selection consistency, which is further illustrated with examples and simulation studies.

math.ST↗

arXiv2025

Tell Me What to Track: Infusing Robust Language Guidance for Enhanced Referring Multi-Object Tracking

Referring multi-object tracking (RMOT) is an emerging cross-modal task that aims to localize an arbitrary number of targets based on a language expression and continuously track them in a video. This intricate task involves reasoning on multi-modal data and precise target localization with temporal association. However, prior studies overlook the imbalanced data distribution between newborn targets and existing targets due to the nature of the task. In addition, they only indirectly fuse multi-modal features, struggling to deliver clear guidance on newborn target detection. To solve the above issues, we conduct a collaborative matching strategy to alleviate the impact of the imbalance, boosting the ability to detect newborn targets while maintaining tracking performance. In the encoder, we integrate and enhance the cross-modal and multi-scale fusion, overcoming the bottlenecks in previous work, where limited multi-modal information is shared and interacted between feature maps. In the decoder, we also develop a referring-infused adaptation that provides explicit referring guidance through the query tokens. The experiments showcase the superior performance of our model (+3.42%) compared to prior works, demonstrating the effectiveness of our designs.

cs.CV↗