Publications

Journal Articles

A novel network for resolving subjective masking differences and accurate thyroid nodule diagnosis Permalink

Published in Computational Biology and Chemistry , 2025

This paper develops a neural network for thyroid nodule diagnosis that explicitly addresses subjective masking differences in clinical annotations. The method is designed to improve feature extraction robustness and deliver more accurate, reliable predictions for ultrasound-based diagnosis.

Recommended citation: Zhiyuan Ouyang. "A novel network for resolving subjective masking differences and accurate thyroid nodule diagnosis." Computational Biology and Chemistry, 2025.
Download Paper

Model detection for grey forecasting model with polynomial term Permalink

Published in Communications in Statistics - Simulation and Computation, 2024

This paper studies grey forecasting models with polynomial terms, covering the traditional grey model, the nonhomogeneous grey model, and integer-order grey models with time-power terms as special cases. It proposes a model detection method for identifying which polynomial orders are truly significant, balancing predictive benefit against added model complexity. Simulations and an empirical study on annual health expenditure in China show that the method can recover the true model structure and deliver strong forecasting performance.

Recommended citation: Zhiyuan Ouyang. "Model detection for grey forecasting model with polynomial term." Communications in Statistics - Simulation and Computation, 2024.
Download Paper

Conference Papers

Primal-Spectral Generative Modeling: Fast Analytical Generation via Pseudoinverse Levy Inversion Permalink

Published in ICML 2026 (CCF-A), 2026

This paper transforms probability distributions into continuous spectral functions to make generative modeling more amenable to neural approximation, and provides theoretical guarantees for convergence to the true distribution. It introduces PriSpecNet together with a 1-NFE Pseudoinverse Levy Inversion solver that reformulates sampling as a fast analytical problem rather than iterative numerical integration. Experiments show strong improvements on time-series generation and forecasting benchmarks, while achieving competitive ImageNet 256x256 generation with FID 1.66 at dramatically lower computational cost.

Recommended citation: Zhiyuan Ouyang. "Primal-Spectral Generative Modeling: Fast Analytical Generation via Pseudoinverse Levy Inversion." ICML 2026.
Download Paper