Chao Zhou

Im Oberen Werk 1,

66386 St. Ingbert

Germany

Hi! I'm Chao (Eddy) Zhou, currently a Postdoctoral Researcher in the Relational Machine Learning Group at the CISPA Helmholtz Center for Information Security, Germany, supervised by Prof. Rebekka Burkholz. I obtain my PhD under the supervision of Prof.Miguel Rodrigues, in Information, Inference and Machine Learning (I2|ML)group at UCL, London, UK, in 2023. I obtained my Bachelor degree in Communication Engineering at WHUT in 2015 and Master Degree in Information and Communication Systems at UESTC in 2018, respectively.

My recent research interests span machine learning, interpretable AI, generative AI, and efficient machine learning.

news

May, 2025	Our work 'Mirror, Mirror of the Flow: How Does Regularization Shape Implicit Bias?' is accepted at ICML 2025.
May, 2024	I am joining the Relational Machine Learning Group at CISPA as a postdoc under the supervision of Dr. Rebekka Burkholz in May 2024.
October, 2023	Our work 'BITS-Net: Blind Image Transparency Separation Network' is presented at ICIP 2023, Kuala Lumpur, Malaysia.
August, 2023	Our work 'Hyperspectral Blind Unmixing using a Double Deep Image Prior' with Prof. Miguel has been accepted at TNNLS 2023.
30 Jun, 2023	I passed my PhD viva with minor corrections on 30 Jun 2023! Thanks to my supervisor Prof. Miguel Rodrigues for arranging this viva. Thanks to internal examiner Prof. Ioannis (Yiannis) Andreopoulos from UCL and external examiner Dr. Wei Dai from Imperial College London.

selected publications

Conference

Mirror, Mirror of the Flow: How Does Regularization Shape Implicit Bias?

Tom Jacobs, C. Zhou, and Rebekka Burkholz

in ICML 2025.

Abs PDF

Implicit bias plays an important role in explaining how overparameterized models generalize well. Explicit regularization like weight decay is often employed in addition to prevent overfitting. While both concepts have been studied separately, in practice, they often act in tandem. Understanding their interplay is key to controlling the shape and strength of implicit bias, as it can be modified by explicit regularization. To this end, we incorporate explicit regularization into the mirror flow framework and analyze its lasting effects on the geometry of the training dynamics, covering three distinct effects: positional bias, type of bias, and range shrinking. Our analytical approach encompasses a broad class of problems, including sparse coding, matrix sensing, single-layer attention, and LoRA, for which we demonstrate the utility of our insights. To exploit the lasting effect of regularization and highlight the potential benefit of dynamic weight decay schedules, we propose to switch off weight decay during training, which can improve generalization, as we demonstrate in experiments.

Journal

Hyperspectral Blind Unmixing Using a Double Deep Image Prior

C. Zhou, and M. R. D. Rodrigues

in IEEE Transactions on Neural Networks and Learning Systems, doi: 10.1109/TNNLS.2023.3294714.

Abs PDF Code

With the rise of machine learning, hyperspectral image (HSI) unmixing problems have been tackled using learning-based methods. However, physically meaningful unmixing results are not guaranteed without proper guidance. In this work, we propose an unsupervised framework inspired by deep image prior (DIP) that can be used for both linear and nonlinear blind unmixing models. The framework consists of three modules: 1) an Endmember estimation module using DIP (EDIP); 2) an Abundance estimation module using DIP (ADIP); and 3) a mixing module (MM). EDIP and ADIP modules generate endmembers and abundances, respectively, while MM produces a reconstruction of the HSI observations based on the postulated unmixing model. We introduce a composite loss function that applies to both linear and nonlinear unmixing models to generate meaningful unmixing results. In addition, we propose an adaptive loss weight strategy for better unmixing results in nonlinear mixing scenarios. The proposed methods outperform state-of-the-art unmixing algorithms in extensive experiments conducted on both synthetic and real datasets.