Texas A&M University

Selected work

Representative Papers

Hashigo: A Next-Generation Sketch Interactive System for Japanese Kanji

Apr 09, 2009Conference on Innovative Applications of Artificial Intelligence

Existing Japanese kanji handwriting recognition systems focus solely on glyph-level matching, neglecting fine-grained assessment of stroke order, writing dynamics, and structural topology—thus failing to correct learners’ erroneous habits. Method: This paper introduces a sketch-based interactive system for Japanese kanji learning, integrating (1) pen-tip kinematic modeling, (2) topological structural analysis, (3) multi-stage stroke-order verification, and (4) rule-guided machine learning—enabling dual-dimensional, high-accuracy automated evaluation of both visual structure and writing technique. Contribution/Results: The system delivers teacher-level real-time feedback, significantly improving writing conformity and recognition accuracy. Empirical evaluation shows 92% agreement between its feedback and that of domain experts, effectively preventing the entrenchment of incorrect writing habits and overcoming fundamental limitations of conventional recognition systems in pedagogical assessment.

35 citations1 influentialRead paper

A Diffusion Model Translator for Efficient Image-to-Image Translation

Jul 30, 2024IEEE Transactions on Pattern Analysis and Machine Intelligence

Existing image-to-image (I2I) translation diffusion models redundantly inject the source image at every denoising step, leading to inefficient inference. This work proposes the lightweight Diffusion Model Translator (DMT), which—through the first theoretical analysis—demonstrates that a single inter-domain distribution transfer suffices for high-fidelity I2I translation. Guided by this insight, DMT introduces a compact translation module that performs distribution alignment only at a critical intermediate timestep. Built upon the DDPM framework, DMT integrates probabilistic distribution shift analysis, an adaptive optimal timestep selection strategy, and a streamlined architecture. Extensive experiments on style transfer, colorization, semantic segmentation map generation, and sketch-to-color tasks show that DMT achieves state-of-the-art performance with significantly faster inference—averaging 3.2× speedup—while simultaneously delivering superior image quality.

5 citationsRead paper

Bayesian Data Sketching for Varying Coefficient Regression Models

May 30, 2025

Bayesian inference for varying-coefficient regression models with large-scale functional data is computationally prohibitive due to the high cost of Markov chain Monte Carlo (MCMC). To address this, we propose a data sketching method based on randomized linear transformations that compresses both the response vector and the predictor matrix into low-dimensional representations, while preserving the full Bayesian modeling framework. Standard MCMC or variational inference tools can then be directly applied to the sketched data. This work marks the first application of Bayesian data sketching to varying-coefficient models—requiring no new model specification, custom algorithm development, or specialized hardware. Theoretically and empirically, the method maintains near-identical statistical efficacy while achieving speedups of several orders of magnitude. Moreover, it seamlessly integrates with existing Bayesian inference ecosystems.

4 citations1 influentialRead paper

Spectral Convolutional Conditional Neural Processes

Apr 19, 2024arXiv.org

Conventional Convolutional Conditional Neural Processes (ConvCNPs) struggle with long-range dependencies in few-shot, irregularly sampled function modeling due to their reliance on local spatial convolutions; increasing kernel size incurs prohibitive computational overhead. Method: This work introduces SpectralCNP—the first integration of Fourier Neural Operator principles into the CNP framework—replacing local spatial convolutions with global spectral-domain convolutions via parameterized Fourier transforms. This enables robust modeling of sparse, unstructured observations. Contribution/Results: Trained via maximum likelihood, SpectralCNP achieves significant improvements over ConvCNP across multiple meta-learning and functional regression benchmarks. Notably, it delivers superior predictive accuracy and calibration—particularly under low-data regimes and irregular sampling—while maintaining computational efficiency through spectral parameterization.

4 citationsRead paper

Audience Amplified: Virtual Audiences in Asynchronously Performed AR Theater

Oct 21, 2024International Symposium on Mixed and Augmented Reality

In mobile augmented reality (AR) settings, single users often lack authentic audience feedback and struggle to achieve social presence. Method: This work proposes an immersive theater experience framework leveraging virtual audiences in mobile AR. It constructs a phygital co-spatial environment enabling real-time interaction between users, AI-driven virtual dancers, and virtual audiences generated via imitation learning. Notably, it pioneers the application of imitation learning to virtual audience modeling—capturing behavioral diversity and imperfection of real crowds while synchronizing audio and motion. Contribution/Results: Experiments demonstrate that virtual audiences significantly enhance users’ perceived sociability and engagement. Embodied virtual avatars are preferred over audio-only feedback. Intriguingly, audience presence exhibits a dual effect: in certain contexts, its absence elicits stronger emotional responses. This study establishes a novel design paradigm and provides empirical evidence for socially immersive asynchronous AR performances.

3 citationsRead paper

Recent publications