arXiv subjects

Cao Yuan

Publications and source records attributed to Cao Yuan.

5 recordsLinked to original sources

arXiv2026

ECTO: Exogenous-Conditioned Temporal Operator for Ultra-Short-Term Wind Power Forecasting

Accurate ultra-short-term wind power forecasting is critical for grid dispatch and reserve management, yet remains challenging due to the non-stationary, condition-dependent nature of wind generation. Meteorological exogenous variables carry substantial predictive information, but the most informative variable combination varies across sites, operating conditions, and prediction horizons. Existing deep learning approaches either treat exogenous inputs as generic auxiliary channels through uniform mixing or soft gating, or rely on fixed preprocessing steps such as PCA, without exploiting the physical structure of meteorological variables. We propose ECTO (Exogenous-Conditioned Temporal Operator), a unified framework that decomposes exogenous variable modeling into two complementary modules. Physically-Grounded Variable Selection (PGVS) performs hierarchical, group-aware sparse selection over exogenous variables using a domain-informed physical prior and sparsemax activations, producing a compact, condition-adaptive exogenous context. Exogenous-Conditioned Regime Refinement (ECRR) routes the forecast through learned regime experts that apply gain--bias calibration and horizon-specific corrections via a mixture-of-experts paradigm. Experiments on three wind farms spanning different climates, capacities (66--200 MW), and exogenous dimensions (11--13 variables) demonstrate that ECTO achieves the lowest MSE across all sites, with relative improvements over the strongest baseline ranging from 2.2% to 5.2%, widening to 8.6% at the longer prediction horizon ($H=32$). Ablation analysis confirms that each exogenous-related component contributes positively (PGVS +1.84%, ECRR +2.86%), and interpretability analysis reveals that PGVS learns physically meaningful, site-specific variable selection patterns, while ECRR converges to well-separated calibration strategies consistent across sites.

cs.LG↗

arXiv2020

Conditional Image Generation with One-Vs-All Classifier

This paper explores conditional image generation with a One-Vs-All classifier based on the Generative Adversarial Networks (GANs). Instead of the real/fake discriminator used in vanilla GANs, we propose to extend the discriminator to a One-Vs-All classifier (GAN-OVA) that can distinguish each input data to its category label. Specifically, we feed certain additional information as conditions to the generator and take the discriminator as a One-Vs-All classifier to identify each conditional category. Our model can be applied to different divergence or distances used to define the objective function, such as Jensen-Shannon divergence and Earth-Mover (or called Wasserstein-1) distance. We evaluate GAN-OVAs on MNIST and CelebA-HQ datasets, and the experimental results show that GAN-OVAs make progress toward stable training over regular conditional GANs. Furthermore, GAN-OVAs effectively accelerate the generation process of different classes and improves generation quality.

cs.CV↗

arXiv2019

A novel method for identifying the deep neural network model with the Serial Number

Deep neural network (DNN) with the state of art performance has emerged as a viable and lucrative business service. However, those impressive performances require a large number of computational resources, which comes at a high cost for the model creators. The necessity for protecting DNN models from illegal reproducing and distribution appears salient now. Recently, trigger-set watermarking, breaking the white-box restriction, relying on adversarial training pre-defined (incorrect) labels for crafted inputs, and subsequently using them to verify the model authenticity, has been the main topic of DNN ownership verification. While these methods have successfully demonstrated robustness against removal attacks, few are effective against the tampering attacks from competitors forging the fake watermarks and dogging in the manager. In this paper, we put forth a new framework of the trigger-set watermark by embedding a unique Serial Number (relatedness less original labels) to the deep neural network for model ownership identification, which is both robust to model pruning and resist to tampering attacks. Experiment results demonstrate that the DNN Serial Number only incurs slight accuracy degradation of the original performance and is valid for ownership verification.

cs.CV↗

arXiv2016

Cyclic codes over $\mathbb{Z}_4[u]/\langle u^k\rangle$ of odd length

Let $R=\mathbb{Z}_{4}[u]/\langle u^k\rangle=\mathbb{Z}_{4}+u\mathbb{Z}_{4}+\ldots+u^{k-1}\mathbb{Z}_{4}$ ($u^k=0$) where $k\in \mathbb{Z}^{+}$ satisfies $k\geq 2$. For any odd positive integer $n$, it is known that cyclic codes over $R$ of length $n$ are identified with ideals of the ring $R[x]/\langle x^{n}-1\rangle$. In this paper, an explicit representation for each cyclic code over $R$ of length $n$ is provided and a formula to count the number of codewords in each code is given. Then a formula to calculate the number of cyclic codes over $R$ of length $n$ is obtained. Precisely, the dual code of each cyclic code and self-dual cyclic codes over $R$ of length $n$ are investigated. When $k=4$, some optimal quasi-cyclic codes over $\mathbb{Z}_{4}$ of length $28$ and index $4$ are obtained from cyclic codes over $R=\mathbb{Z}_{4} [u]/\langle u^4\rangle$.

cs.IT↗

arXiv2016

On a class of left metacyclic codes

Let $G_{(m,3,r)}=\langle x,y\mid x^m=1, y^3=1,yx=x^ry\rangle$ be a metacyclic group of order $3m$, where ${\rm gcd}(m,r)=1$, $1<r<m$ and $r^3\equiv 1$ (mod $m$). Then left ideals of the group algebra $\mathbb{F}_q[G_{(m,3,r)}]$ are called left metacyclic codes over $\mathbb{F}_q$ of length $3m$, and abbreviated as left $G_{(m,3,r)}$-codes. A system theory for left $G_{(m,3,r)}$-codes is developed for the case of ${\rm gcd}(m,q)=1$ and $r\equiv q^ε$ for some positive integer $ε$, only using finite field theory and basic theory of cyclic codes and skew cyclic codes. The fact that any left $G_{(m,3,r)}$-code is a direct sum of concatenated codes with inner codes ${\cal A}_i$ and outer codes $C_i$ is proved, where ${\cal A}_i$ is a minimal cyclic code over $\mathbb{F}_q$ of length $m$ and $C_i$ is a skew cyclic code of length $3$ over an extension field of $\mathbb{F}_q$. Then an explicit expression for each outer code in any concatenated code is provided. Moreover, the dual code of each left $G_{(m,3,r)}$-code is given and self-orthogonal left $G_{(m,3,r)}$-codes are determined.

cs.IT↗