arXiv subjects

Xiang Meng

Publications and source records attributed to Xiang Meng.

At least 19 recordsLinked to original sources

arXiv2026

Average Cause-Specific Hazard: A Censoring-Invariant Measure of Event Burden Under Competing Risks

Competing events are common in clinical and epidemiologic studies, including semi-competing risks in which a terminal event such as death may follow a nonfatal event but also competes with it beforehand. Standard summaries include the cumulative incidence function (CIF) and the incidence rate (IR), defined as the number of observed events divided by observed event-free person-time. With competing events, the naive IR generally depends on the censoring-time distribution unless intensities are constant. We propose the Average Cause-Specific Hazard (ACSH), a survival-weighted rate per event-free person-time that preserves the interpretation of an incidence rate and is defined purely from the event-time distribution, without involving the censoring-time distribution. We develop nonparametric estimation and inference for ACSH and, for two-sample comparisons, introduce ACSH differences and ratios that provide interpretable contrasts without requiring a strong model assumption between two groups. Simulation studies examine the finite-sample performance, and an analysis of the CANVAS trial illustrates the proposed methods.

stat.ME↗

arXiv2026

ADMM-Q: An Improved Hessian-based Weight Quantizer for Post-Training Quantization of Large Language Models

Quantization is an effective strategy to reduce the storage and computation footprint of large language models (LLMs). Post-training quantization (PTQ) is a leading approach for compressing LLMs. Popular weight quantization procedures, including GPTQ and RTN, suffer in model utility, especially at aggressive quantization levels (sub-4-bit). We propose ADMM-Q, a novel weight quantization algorithm that considers the layer-wise quantization problem. Our algorithm is based on a combinatorial variant of the Alternating Direction Method of Multipliers (ADMM). Our operator-splitting procedure updates weights continuously to minimize the layer-wise reconstruction error, while gradually enforcing the quantization constraints with convergence guarantees. We propose additional algorithmic enhancements (e.g., penalty scheduling, preconditioning, and a local search post-processing step) to make ADMM-Q efficient at LLM scale. ADMM-Q is modular and can be used as a drop-in replacement for any weight quantizer within existing quantization pipelines: ADMM-Q is fully composable with existing techniques including range clipping, learned or random rotations, and activation scaling. Using ADMM-Q in place of GPTQ on Qwen3-8B, we decrease WikiText-2 perplexity in: (i) the W3A16 weight-only setting (12.85 $\rightarrow$ 10.06); (ii) the W4A8 SmoothQuant procedure (9.29 $\rightarrow$ 8.68); and (iii) the W2A4KV4 SpinQuant procedure (66.11 $\rightarrow$ 19.42).

cs.LG↗

arXiv2026

MOONSHOT : A Framework for Multi-Objective Pruning of Vision and Large Language Models

Weight pruning is a common technique for compressing large neural networks. We focus on the challenging post-training one-shot setting, where a pre-trained model is compressed without any retraining. Existing one-shot pruning methods typically optimize a single objective, such as a layer-wise reconstruction loss or a second-order Taylor approximation of the training loss. We highlight that neither objective alone is consistently the most effective across architectures and sparsity levels. Motivated by this insight, we propose MOONSHOT, a general and flexible framework that extends any single-objective pruning method into a multi-objective formulation by jointly optimizing both the layer-wise reconstruction error and second-order Taylor approximation of the training loss. MOONSHOT acts as a wrapper around existing pruning algorithms. To enable this integration while maintaining scalability to billion-parameter models, we propose modeling decisions and introduce an efficient procedure for computing the inverse Hessian, preserving the efficiency of state-of-the-art one-shot pruners. When combined with state-of-the-art pruning methods on Llama-3.2 and Llama-2 models, MOONSHOT reduces C4 perplexity by up to 32.6% at 2:4 sparsity and improves zero-shot mean accuracy across seven classification benchmarks by up to 4.9 points. On Vision Transformers, it improves accuracy on ImageNet-1k by over 5 points at 70% sparsity, and on ResNet-50, it yields a 4-point gain at 90% sparsity.

cs.LG↗

arXiv2026

Computation of Least Trimmed Squares: A Branch-and-Bound framework with Hyperplane Arrangement Enhancements

We study computational aspects of a key problem in robust statistics -- the penalized least trimmed squares (LTS) regression problem, a robust estimator that mitigates the influence of outliers in data by capping residuals with large magnitudes. Although statistically attractive, penalized LTS is NP-hard, and existing mixed-integer optimization (MIO) formulations scale poorly due to weak relaxations and exponential worst-case complexity in the number of observations. We propose a new MIO formulation that embeds hyperplane arrangement logic into a perspective reformulation, explicitly enforcing structural properties of optimal solutions. We show that, if the number of features is fixed, the resulting branch-and-bound tree is of polynomial size in the sample size. Moreover, we develop a tailored branch-and-bound algorithm that uses first-order methods with dual bounds to solve node relaxations efficiently. Computational experiments on synthetic and real datasets demonstrate substantial improvements over existing MIO approaches: on synthetic instances with 5000 samples and 20 features, our tailored solver reaches a 1% gap in 1 minute while competing approaches fail to do so within one hour. These gains enable exact robust regression at significantly larger sample sizes in low-dimensional settings.

math.OC↗

arXiv2026

CellE: Automated Standard Cell Library Extension via Equality Saturation

Automated standard cell library extension is crucial for maximizing Quality of Results (QoR) in modern VLSI design. We introduce CellE, a novel framework that leverages formal methods to achieve exhaustive discovery of functionally equivalent subcircuits. CellE applies equality saturation to the post-mapping netlist, generating an e-graph to cluster all functionally equivalent implementations. This canonical representation enables an efficient pattern mining algorithm to select the most area-optimal standard cells. Experimental results show a 15.41% average area reduction (up to 23.64% over prior work). Furthermore, characterization in a commercial flow demonstrates an 8.00% average delay reduction, confirming CellE's superior QoR optimization capabilities.

cs.AR↗

arXiv2026

3BASiL: An Algorithmic Framework for Sparse plus Low-Rank Compression of LLMs

Sparse plus Low-Rank $(\mathbf{S} + \mathbf{LR})$ decomposition of Large Language Models (LLMs) has emerged as a promising direction in model compression, aiming to decompose pre-trained model weights into a sum of sparse and low-rank matrices $(\mathbf{W} \approx \mathbf{S} + \mathbf{LR})$. Despite recent progress, existing methods often suffer from substantial performance degradation compared to dense models. In this work, we introduce 3BASiL-TM, an efficient one-shot post-training method for $(\mathbf{S} + \mathbf{LR})$ decomposition of LLMs that addresses this gap. Our approach first introduces a novel 3-Block Alternating Direction Method of Multipliers (ADMM) method, termed 3BASiL, to minimize the layer-wise reconstruction error with convergence guarantees. We then design an efficient transformer-matching (TM) refinement step that jointly optimizes the sparse and low-rank components across transformer layers. This step minimizes a novel memory-efficient loss that aligns outputs at the transformer level. Notably, the TM procedure is universal as it can enhance any $(\mathbf{S} + \mathbf{LR})$ decomposition, including pure sparsity. Our numerical experiments show that 3BASiL-TM reduces the WikiText2 perplexity gap relative to dense LLaMA-8B model by over 30% under a (2:4 Sparse + 64 LR) configuration, compared to prior methods. Moreover, our method achieves over 2.5x faster compression runtime on an A100 GPU compared to SOTA $(\mathbf{S} + \mathbf{LR})$ method. Our code is available at https://github.com/mazumder-lab/3BASiL.

cs.LG↗

arXiv2026

Efficient and Debiased Learning of Average Hazard Under Non-Proportional Hazards

The hazard ratio from the Cox proportional hazards model is a ubiquitous summary of treatment effect. However, when hazards are non-proportional, the hazard ratio can lose a stable causal interpretation and become study-dependent because it effectively averages time-varying effects with weights determined by follow-up and censoring. We consider the average hazard (AH) as an alternative causal estimand: a population-level person-time event rate that remains well-defined and interpretable without assuming proportional hazards. Although AH can be estimated nonparametrically and regression-style adjustments have been proposed, existing approaches do not provide a general framework for flexible, high-dimensional nuisance estimation with valid sqrt{n} inference. We address this gap by developing a semiparametric, doubly robust framework for covariate-adjusted AH. We establish pathwise differentiability of AH in the nonparametric model, derive its efficient influence function, and construct cross-fitted, debiased estimators that leverage machine learning for nuisance estimation while retaining asymptotically normal, sqrt{n}-consistent inference under mild product-rate conditions. Simulations demonstrate that the proposed estimator achieves small bias and near-nominal confidence-interval coverage across proportional and non-proportional hazards settings, including crossing-hazards regimes where Cox-based summaries can be unstable. We illustrate practical utility in comparative effectiveness research by comparing immunotherapy regimens for advanced melanoma using SEER-Medicare linked data.

stat.ME↗

arXiv2026

A GPU-accelerated Nonlinear Branch-and-Bound Framework for Sparse Linear Models

We study exact sparse linear regression with an $\ell_0-\ell_2$ penalty and develop a branch-and-bound (BnB) algorithm explicitly designed for GPU execution. Starting from a perspective reformulation, we derive an interval relaxation that can be solved by ADMM with closed-form, coordinate-wise updates. We structure these updates so that the main work at each BnB node reduces to batched matrix-vector operations with a shared data matrix, enabling fine-grained parallelism across coordinates and coarse-grained parallelism across many BnB nodes on a single GPU. Feasible solutions (upper bounds) are generated by a projected gradient method on the active support, implemented in a batched fashion so that many candidate supports are updated in parallel on the GPU. We discuss practical design choices such as memory layout, batching strategies, and load balancing across nodes that are crucial for obtaining good utilization on modern GPUs. On synthetic and real high-dimensional datasets, our GPU-based approach achieves clear runtime improvements over a CPU implementation of our method, an existing specialized BnB method, and commercial MIP solvers.

math.OC↗

arXiv2025

Multimode magnon-phonon cavity driven by symmetry-locked strain fields

Hybrid magnon-phonon cavities with precise control knobs are highly sought after for coherent energy and signal transduction in solid-state platforms. While strain offers a powerful means to tune magnonic characteristics, extending strain engineering into magnon-phonon hybridization has remained elusive. Moreover, implementing controllable strain at the meso- or nanoscale poses a formidable challenge, as spatial inhomogeneity of strain fields often leads to enhanced damping and reduced coherence, thereby hindering device integration and scalability. Here, we present an epitaxial La0.7Sr0.3MnO3/SrTiO3 (LSMO/STO) heterostructure that exhibits strong coupling between the Kittel magnon and acoustic phonon. Leveraging the emergence of structural domains when STO undergoes a cubic-to-tetragonal phase transition, we create anisotropic local strains at the interface. Remarkably, the anisotropic local strain of less than 0.1 % drives the pronounced splitting of the magnon into three branches. Each branch independently hybridizes with acoustic phonons, forming a matrix of magnon-phonon avoided crossings that underpins multimode transduction and programmable networks in frequency and magnetic field space. An analytical model reveals that the split magnon branches are deterministically locked to the three main crystalline axes, enabling robust, orientation-selective control of the hybridized magnon-phonon spectrum against spatial inhomogeneity. Our results establish designed local strain as an exceptionally sensitive trigger for multimode magnon-phonon hybridization in magnetoelastic oxide heterostructures, and highlight local strain engineering as a viable strategy for designing tunable hybrid magnonic and phononic devices.

cond-mat.mtrl-sci↗

arXiv2025

SpeechForensics: Audio-Visual Speech Representation Learning for Face Forgery Detection

Detection of face forgery videos remains a formidable challenge in the field of digital forensics, especially the generalization to unseen datasets and common perturbations. In this paper, we tackle this issue by leveraging the synergy between audio and visual speech elements, embarking on a novel approach through audio-visual speech representation learning. Our work is motivated by the finding that audio signals, enriched with speech content, can provide precise information effectively reflecting facial movements. To this end, we first learn precise audio-visual speech representations on real videos via a self-supervised masked prediction task, which encodes both local and global semantic information simultaneously. Then, the derived model is directly transferred to the forgery detection task. Extensive experiments demonstrate that our method outperforms the state-of-the-art methods in terms of cross-dataset generalization and robustness, without the participation of any fake video in model training. Code is available at https://github.com/Eleven4AI/SpeechForensics.

cs.CV↗

arXiv2025

Variance estimation after matching or re-weighting

This paper develops a variance estimation framework for matching estimators that enables valid population inference for treatment effects. We provide theoretical analysis of a variance estimator that addresses key limitations in the existing literature. While Abadie and Imbens (2006) proposed a foundational variance estimator requiring matching for both treatment and control groups, this approach is computationally prohibitive and rarely used in practice. Our method provides a computationally feasible alternative that only requires matching treated units to controls while maintaining theoretical validity for population inference. We make three main contributions. First, we establish consistency and asymptotic normality for our variance estimator, proving its validity for average treatment effect on the treated (ATT) estimation in settings with small treated samples. Second, we develop a generalized theoretical framework with novel regularity conditions that significantly expand the class of matching procedures for which valid inference is available, including radius matching, M-nearest neighbor matching, and propensity score matching. Third, we demonstrate that our approach extends naturally to other causal inference estimators such as stable balancing weighting methods. Through simulation studies across different data generating processes, we show that our estimator maintains proper coverage rates while the state-of-the-art bootstrap method can exhibit substantial undercoverage (dropping from 95% to as low as 61%), particularly in settings with extensive control unit reuse. Our framework provides researchers with both theoretical guarantees and practical tools for conducting valid population inference across a wide range of causal inference applications. An R package implementing our method is available at https://github.com/jche/scmatch2.

stat.ME↗

arXiv2025

TSENOR: Highly-Efficient Algorithm for Finding Transposable N:M Sparse Masks

Network pruning reduces the computational requirements of large neural networks, with N:M sparsity -- retaining only N out of every M consecutive weights -- offering a compelling balance between compressed model quality and hardware acceleration. However, N:M sparsity only accelerates forward-pass computations, as N:M patterns are not preserved during matrix transposition, limiting efficiency during training where both passes are computationally intensive. While transposable N:M sparsity has been proposed to address this limitation, existing methods for finding transposable N:M sparse masks either fail to scale to large models or are restricted to M=4 which results in suboptimal compression-accuracy trade-off. We introduce an efficient solver for transposable N:M masks that scales to billion-parameter models. We formulate mask generation as optimal transport problems and solve through entropy regularization and Dykstra's algorithm, followed by a rounding procedure. Our tensor-based implementation exploits GPU parallelism, achieving up to 100x speedup with only 1-10% error compared to existing methods. Our approach can be integrated with layer-wise N:M pruning frameworks including Wanda, SparseGPT and ALPS to produce transposable N:M sparse models with arbitrary N:M values. Experiments show that LLaMA3.2-8B with transposable 16:32 sparsity maintains performance close to its standard N:M counterpart and outperforms standard 2:4 sparse model, showing the practical value of our approach.

cs.LG↗

arXiv2025

Highly Uniform Thermally Undercut Silicon Photonic Devices in a 300 mm CMOS Foundry Process

Silicon photonic devices fundamental to high-density wavelength-division multiplexed (DWDM) optical links and photonic switching networks, such as resonant modulators and Mach-Zehnder interferometers (MZIs), are highly sensitive to fabrication variations and operational temperature swings. However, thermal tuning to compensate for fabrication and operational temperature variations can result in prohibitive power consumption, challenging the scalability of energy-efficient photonic integrated circuits (PICs). In this work, we develop and demonstrate a wafer-scale thermal undercut process in a 300 mm complementary metal oxide semiconductor (CMOS) foundry that dramatically improves the thermal isolation of thermo-optic devices by selectively removing substrate material beneath the waveguides and resonators. This approach significantly reduces the power required for thermal tuning across multiple device architectures, achieving almost a 5$\times$ improvement in tuning efficiency in a state-of-the-art 4.5 $\mu$m radius microdisk modulator and a 40$\times$ improvement in efficiency for a MZI phase shifter. To the best of the authors' knowledge, we demonstrate the first wafer-scale comparison of non-undercut and undercut silicon photonic devices using comprehensive wafer-scale measurements across 64 reticles of a 300 mm silicon-on-insulator (SOI) wafer. Further, we demonstrate a comprehensive wafer-scale analysis of the influence of undercut trench opening geometry on device tuning efficiency. Notably, we observe highly uniform performance across the full 300 mm wafer for multiple device types, emphasizing that our process can be scaled to large-scale photonic circuits with high yield. These results open new opportunities for large-scale integrated photonic circuits using thermo-optic devices, paving the way for scalable, low-power silicon photonic systems.

physics.optics↗

arXiv2025

Time-Varying Causal Survival Learning

This work bridges the gap between staggered adoption designs and survival analysis to estimate causal effects in settings with time-varying treatments, addressing a fundamental challenge in medical research exemplified by the Stanford Heart Transplant study. In medical interventions, particularly organ transplantation, the timing of treatment varies significantly across patients due to factors such as donor availability and patient readiness, introducing potential bias in treatment effect estimation if not properly accounted for. We identify conditions under which staggered adoption assumptions can justify the use of survival analysis techniques for causal inference with time-varying treatments. By establishing this connection, we enable the use of existing survival analysis methods while maintaining causal interpretability. Furthermore, we enhance estimation performance by incorporating double machine learning methods, improving efficiency when handling complex relationships between patient characteristics and survival outcomes. Through both simulation studies and application to heart transplant data, our approach demonstrates superior performance compared to traditional methods, reducing bias and offering theoretical guarantees for improved efficiency in survival analysis settings.

stat.ME↗

arXiv2025

Evaluation of the HeartSteps Online Sampling Algorithm

Micro-randomized trials (MRTs), which sequentially randomize participants at multiple decision times, have gained prominence in digital intervention development. These sequential randomizations are often subject to certain constraints. In the MRT called HeartSteps V2V3, where an intervention is designed to interrupt sedentary behavior, two core design constraints need to be managed: an average of 1.5 interventions across days and the uniform delivery of interventions across decision times. Meeting both constraints, especially when the times allowed for randomization are not determined beforehand, is challenging. An online algorithm was implemented to meet these constraints in the HeartSteps V2V3 MRT. We present a case study using data from the HeartSteps V2V3 MRT, where we select appropriate metrics, discuss issues in making an accurate evaluation, and assess the algorithm's performance. Our evaluation shows that the algorithm performed well in meeting the two constraints. Furthermore, we identify areas for improvement and provide recommendations for designers of MRTs that need to satisfy these core design constraints.

stat.AP↗

arXiv2024

Integrating Chaotic Evolutionary and Local Search Techniques in Decision Space for Enhanced Evolutionary Multi-Objective Optimization

This paper presents innovative approaches to optimization problems, focusing on both Single-Objective Multi-Modal Optimization (SOMMOP) and Multi-Objective Optimization (MOO). In SOMMOP, we integrate chaotic evolution with niching techniques, as well as Persistence-Based Clustering combined with Gaussian mutation. The proposed algorithms, Chaotic Evolution with Deterministic Crowding (CEDC) and Chaotic Evolution with Clustering Algorithm (CECA), utilize chaotic dynamics to enhance population diversity and improve search efficiency. For MOO, we extend these methods into a comprehensive framework that incorporates Uncertainty-Based Selection, Adaptive Parameter Tuning, and introduces a radius $ R $ concept in deterministic crowding, which enables clearer and more precise separation of populations at peak points. Experimental results demonstrate that the proposed algorithms outperform traditional methods, achieving superior optimization accuracy and robustness across a variety of benchmark functions.

cs.NE↗

arXiv2024

Caliper Synthetic Matching: Generalized Radius Matching with Local Synthetic Controls

Matching promises transparent causal inferences for observational data, making it an intuitive approach for many applications. In practice, however, standard matching methods often perform poorly compared to modern approaches such as response-surface modeling and optimizing balancing weights. We propose Caliper Synthetic Matching (CSM) to address these challenges while preserving simple and transparent matches and match diagnostics. CSM extends Coarsened Exact Matching by incorporating general distance metrics, adaptive calipers, and locally constructed synthetic controls. We show that CSM can be viewed as a monotonic imbalance bounding matching method, so that it inherits the usual bounds on imbalance and bias enjoyed by MIB methods. We further provide a bound on a measure of joint covariate imbalance. Using a simulation study, we illustrate how CSM can even outperform modern matching methods in certain settings, and finally illustrate its use in an empirical example. Overall, we find CSM allows for many of the benefits of matching while avoiding some of the costs.

stat.ME↗

arXiv2024

GLIMMER: Incorporating Graph and Lexical Features in Unsupervised Multi-Document Summarization

Pre-trained language models are increasingly being used in multi-document summarization tasks. However, these models need large-scale corpora for pre-training and are domain-dependent. Other non-neural unsupervised summarization approaches mostly rely on key sentence extraction, which can lead to information loss. To address these challenges, we propose a lightweight yet effective unsupervised approach called GLIMMER: a Graph and LexIcal features based unsupervised Multi-docuMEnt summaRization approach. It first constructs a sentence graph from the source documents, then automatically identifies semantic clusters by mining low-level features from raw texts, thereby improving intra-cluster correlation and the fluency of generated sentences. Finally, it summarizes clusters into natural sentences. Experiments conducted on Multi-News, Multi-XScience and DUC-2004 demonstrate that our approach outperforms existing unsupervised approaches. Furthermore, it surpasses state-of-the-art pre-trained multi-document summarization models (e.g. PEGASUS and PRIMERA) under zero-shot settings in terms of ROUGE scores. Additionally, human evaluations indicate that summaries generated by GLIMMER achieve high readability and informativeness scores. Our code is available at https://github.com/Oswald1997/GLIMMER.

cs.CL↗