Yongxia Chen

Yongxia Chen

Auto-generated profile, based on publicly available information

Yongxia Chen

School of Cyber Science and Engineering, Zhengzhou University, Zhengzhou, Henan, China

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 December 2022
Verified
Conference Paper
Open access
CYCong Yang
Cong Yang
ZLZiliang Li
Ziliang Li
HFHaoyi Fan
Haoyi Fan
YCYongxia Chen
Yongxia Chen
NLNan Lin
Nan Lin
NWNa Wei
Na Wei
Medical report generation is a rising application of big data, which mitigates burden on doctors in clinical trail. However, the serious imbalanced distribution of diseases leads to data bias, an important issue in qualified medical reports generation. To address this issue, a sequence-to-sequence model with incorporation of clinical experience is proposed to generate Chinese ElectroCardioGraph (ECG) reports. The proposed model consists of Electrocardiograph Feature Extractor (EFE), Posterior Knowledge Embedding (PKE) and Report Generator (RG). Firstly, we introduce a novel spatial-temporal information fusion module in EFE to extract robust features from ECG data. Then, embeddings of ECG tags extracted from clinical ECG reports combined with output of EFE are then feed into PKE, which builds a bridge between ECG tags and ECG data, alleviating the problem caused by data bias. Finally, a transformer-based decoder is used in RG to generate ECG reports step by step with output of PKE as key and value. Experiments conducted on private data show that the proposed model can obtain an accuracy 50.26% on BLEU-4, 7.69% higher than state-of-the-art. Our method can also achieve better fluent reports, as demonstrated by the performance on CIDEr, a commonly used content metric.

22 December 2024
Verified
Conference Paper
Open access
CYCong Yang
Cong Yang
QZQianyi Zhai
Qianyi Zhai
YCYongxia Chen
Yongxia Chen
In the field of resource allocation, time series forecasting is crucial for optimizing resource distribution and improving operational efficiency. The core challenge lies in accurately capturing changes in the input sequence to ensure highly accurate forecasting results. Existing methods typically first capture local features of the input sequence, followed by capturing global features using a self-attention mechanism for forecasting. However, these methods suffer from high computational complexity and insufficient feature extraction. In complex scenarios, these shortcomings significantly impact forecasting results. To address this issue, this paper proposes the STLNet (Seasonal-Trend-LongTerm Forecasting Network) model, aiming to deeply capture both trend and seasonal features of sequences, achieving higher forecasting accuracy while reducing computational complexity. Specifically, STLNet combines the extraction of trend and seasonal features and introduces an interval information forecasting layer with stochastic pooling for trend forecasting, enabling more accurate forecasting of trend data. Additionally, for seasonal data forecasting, the model not only considers the correlations within channels but also captures inter-channel correlations through a channel interaction convolution layer. We tested STLNet on several public datasets, and the results show that the model significantly outperforms state-of-the-art methods in terms of accuracy.

22 December 2024
Verified
Conference Paper
Open access
NLNan Lin
Nan Lin
LWLihao Wang
Lihao Wang
CYCong Yang
Cong Yang
YHYing Han
Ying Han
YCYongxia Chen
Yongxia Chen
The issue of catastrophic forgetting is pervasive in class incremental learning scenarios, where the loss of old task knowledge significantly impacts model performance. Traditional knowledge distillation methods often struggle to effectively extract such knowledge. To tackle this challenge, this paper introduces a class incremental learning approach based on multi granularity knowledge distillation. This method regulates the training of the incremental model by extracting detailed and abstract features of the old model's knowledge. It encourages the incremental model to align with the outputs of the old model, thereby retaining more knowledge from previous tasks and preventing catastrophic forgetting. The proposed method can be integrated into various frameworks of class incremental learning. Comparative experiments are conducted using the CIFAR-100 and CIFAR-10 datasets. The results demonstrate that the proposed method effectively mitigates catastrophic forgetting and enhances the classification performance of incremental learning algorithms.