Institution profile

Pennsylvania State University

Academic institutionnorthamerica · us
Official website
Research library1,104linked papers
Opportunities0open roles
Selected work

Representative Papers

Projection Inference for set-identified SVARs

Apr 18, 2025

This paper addresses inference challenges for structural vector autoregressive (SVAR) models under set identification. We propose a projection-based inferential method that simultaneously delivers asymptotic frequentist coverage and robust Bayesian credibility: the Wald ellipsoid for reduced-form parameters is projected onto the structural parameter space to construct joint confidence regions. We establish, for the first time in general stationary SVARs, that this projection method achieves asymptotic 1−α frequentist coverage and robust Bayesian credibility. Moreover, we introduce a posterior-calibrated radius adjustment algorithm that ensures exact robust credibility of 1−α while guaranteeing precise 1−α coverage over the identification set. Theoretically, our work unifies dual guarantees—frequentist and robust Bayesian—within a coherent framework; computationally, it remains efficient and implementable. Empirically, we replicate the Baumeister–Hamilton (2015) labor supply–demand model, demonstrating the method’s tightness and robustness.

15 citations1 influentialRead paper

Deeper or Wider: A Perspective from Optimal Generalization Error with Sobolev Loss

Jan 31, 2024International Conference on Machine Learning

This work investigates the optimal generalization error trade-off between deep neural networks (DeNNs) and wide neural networks (WeNNs) under Sobolev-norm losses. Addressing the “depth vs. width” architectural selection problem, we establish, for the first time, a theoretically grounded criterion based on Sobolev regularity: wide networks dominate under high parameter budgets, whereas deep architectures excel with large sample sizes and higher-order Sobolev loss regularization. Methodologically, we integrate Sobolev-space generalization error bounds with the Deep Ritz method and the physics-informed neural network (PINN) framework to derive interpretable, theory-driven design principles. These principles are empirically validated on PDE-solving tasks using both Deep Ritz and PINN approaches. Our results provide the first generalization-error-theoretic foundation for depth–width selection in PDE numerical solvers, bridging theoretical learning guarantees with practical neural PDE discretization.

12 citations1 influentialRead paper

FLTrojan: Privacy Leakage Attacks against Federated Language Models Through Selective Weight Tampering

Oct 24, 2023arXiv.org

This work uncovers a novel targeted privacy leakage threat against federated language models (FLMs) in federated learning (FL): adversaries—via malicious clients—selectively perturb critical model weights during intermediate training rounds, enabling efficient extraction of others’ sensitive data without server cooperation. Unlike existing general-purpose data extraction attacks, we are the first to identify that intermediate model snapshots exhibit higher privacy vulnerability than final models, and we introduce a principled, weight-sensitivity-driven paradigm for targeted weight manipulation. Integrating membership inference, gradient analysis, and privacy-preserving data reconstruction modeling, our approach achieves a 29% improvement in membership inference recall and reconstructs private data with 71% accuracy—substantially outperforming strong-assumption baseline attacks.

7 citationsRead paper

MoE Lens -- An Expert Is All You Need

Mar 06, 2026

This work investigates the opaque expert specialization mechanism in Mixture-of-Experts (MoE) models, which limits inference and memory efficiency. By analyzing domain-specific routing patterns and employing an early-decoding framework, the study systematically examines how individual experts contribute to model outputs. Through comprehensive analyses—including routing distribution statistics, cosine similarity of hidden states, comparisons between single-expert and ensemble outputs, and perplexity evaluation—the authors find that a small subset of experts handles over 50% of all requests. Remarkably, outputs from a single dominant expert exhibit high consistency with the full model (cosine similarity up to 0.95), with only a 5% increase in perplexity. These findings suggest that precise expert pruning can substantially enhance inference efficiency without compromising performance, offering a promising avenue for efficient MoE deployment and knowledge localization.

5 citations1 influentialRead paper

An Optimal and Scalable Matrix Mechanism for Noisy Marginals under Convex Loss Functions

May 14, 2023Neural Information Processing Systems

Existing methods (e.g., HDMM) for releasing marginal queries over high-dimensional data (up to 100 dimensions) under differential privacy—especially for composite workloads combining marginals with range or prefix-sum queries—suffer from memory explosion and computational intractability. Method: We propose an efficient, unbiased Gaussian noise matrix mechanism that enables global optimization of arbitrary loss objectives expressible as convex functions of marginal variances. Our approach integrates residual planning (ResidualPlanner), convex optimization modeling, sparse linear algebra acceleration, and analytical computation of variance–covariance matrices. Results: Experiments demonstrate scalability: optimizing tens of thousands of marginals completes in seconds; hundred-attribute datasets are processed within two minutes. Memory consumption is reduced by one to two orders of magnitude. Crucially, our method is the first to support exact per-marginal variance and covariance output at scale—enabling principled downstream analysis and adaptive query answering in large-scale differentially private data release.

4 citationsRead paper
Recent publications

Latest Papers