Projects & Related Publications

🎭 Tracing Partisan Bias to Its Emotional Fingerprints: A Computational Approach to Mitigation

IEEE International Conference on Big Data, 2025

This study introduces a novel framework for analyzing and mitigating media bias by tracing partisan stances to their linguistic roots in emotional language. We posit that partisan bias is not merely an abstract stance but materializes as quantifiable 'emotional fingerprints' within news texts. These fingerprints are systematically measured using the Valence-Arousal-Dominance (VAD) framework, allowing us to decode the affective strategies behind partisan framing. Our analysis of the Allsides dataset confirms this hypothesis, revealing distinct and statistically significant emotional fingerprints for left, centre, and right-leaning media. Based on this evidence-driven approach, we then propose a computational approach to mitigation through NeutraSum, a model designed to neutralise these identified emotional patterns. By explicitly targeting the VAD characteristics of biased language, NeutraSum generates summaries that are not only coherent but also demonstrably closer to an emotionally neutral baseline. Experimental results validate our framework: NeutraSum successfully erases the partisan emotional fingerprints from its summaries, achieving a demonstrably lower emotional bias score than other models. This work pioneers a new path for bias mitigation, shifting the focus from treating symptoms (political labels) to addressing the cause: the emotional encoding of partisan bias in language.
We will update the Arxiv version as soon as possible

unjie Liu, Xi Luo, Sirong Wu, Gengchen Sun, Yuhui Deng

Knowledge Distillation with Adapted Weight

Statistics, 2025

We propose the Knowledge Distillation with Adaptive Influence Weight (KD-AIF) framework which leverages influence functions from robust statistics to assign weights to training data, rounded in the four key SAFE principles: Sustainability, Accuracy, Fairness, and Explainability. This novel approach not only optimizes distillation but also increases transparency by revealing the significance of different data. The exploration of various update mechanisms within the KD-AIF framework further elucidates its potential to significantly improve learning efficiency and generalization in student models, marking a step toward more explainable and deployable Large Models.

Sirong Wu, Xi Luo, Junjie Liu, Yuhui Deng

VSG-GAN: A High Fidelity Image Synthesis Method with Semantic Manipulation in Retinal Fundus Image

Biophysical Journal, 2024

In VSG-GAN, a generator and discriminator engage in an adversarial process to produce realistic retinal images. Our approach decouples retinal image generation into distinct modules: the vascular skeleton and background style. Leveraging style transformation and GAN inversion, our proposed hierarchical variational autoencoder module generates retinal images with diverse morphological traits. In addition, the spatially adaptive denormalization module ensures consistency between input and generated images.

Junjie Liu, Shixin Xu, Ping He, Sirong Wu, Xi Luo, Yuhui Deng, Huaxiong Huang

Online Statistics Teaching-assisted Platform with Interactive Web Applications using R Shiny

The 6th International Symposium on Emerging Technologies for Education, 2021

The study of uncertainty is one of the essential parts of statistics, but not easy for students to understand especially in elementary statistical classes. With the rise of new technologies and media, it is worthwhile to think about how to promote innovation in class teaching combining these new technologies with online platforms. In this article, we develop a collection of dynamic interactive web-based applications with Shiny package based on our textbook “Modern Elementary Statistics”.

Junjie Liu, Yuhui Deng, Xiaoling Peng

Analysis of main risk factors causing stroke in Shanxi Province based on machine learning models

Informatics in Medicine Unlocked, 2021

The study analyzes stroke risk factors in Shanxi Province using machine learning models on community and hospital data. Hypertension, physical inactivity, and overweight are identified as key factors. Random forest and logistic regression quantify stroke risks, providing insights for prevention. Results emphasize geographic and lifestyle-specific variations in stroke risks.

Junjie Liu, Yiyang Sun, Jing Ma, Jiachen Tu, Yuhui Deng, Ping He, Rongshan Li, Fengyun Hu Huaxiong Huang, Xiaoshuang Zhou, Shixin Xu

A variant RSA acceleration with parallelisation

International Journal of Parallel, Emergent and Distributed Systems, 2021

The standard RSA relies on multiple big-number modular exponentiation operations and a longer key-length is required for better protection. This imposes a hefty time penalty for encryption and decryption. In this study, we analysed and developed an improved parallel algorithm (PMKRSA) based on the idea of splitting the plaintext into multiple chunks and encrypt the chunks using multiple key-pairs. The algorithm in our new scheme is so natural for parallelised implementation that we also investigated its parallelisation in a GPU environment. In the following, the structure of our new scheme is outlined and its correctness is proved mathematically. Then, with the algorithm implemented and optimised on both CPU and CPU+GPU platforms, we showed that our algorithm shortens the computational time considerably, and it has a security advantage over the standard RSA as it is invulnerable to the common attacks. Finally, we also proved the feasibility of using our algorithm to encrypt large files through simulation. The results show that over the set of file size: 1 MB, 10 MB, 25 MB, 50 MB, 100 MB, the average encryption and decryption time of the CPU version is 0.2476 and 9.4476 s, and for the CPU+GPU version, it is 0.0009 and 0.0618 s, respectively.

Junjie Liu, Kang-Too Tsang, Yuhui Deng