Hello Yu, I’m glad to meet you! Where are you from?

I am originally from Jiangsu Province, China. Currently, I live in Dortmund, Germany, where I work at the Leibniz Institute for Analytical Sciences (ISAS).

What is your background and your professional activity?

My professional background is in biomedical image processing, where I focus on applying AI algorithms to analyze and enhance imaging data. A key aspect of my work is improving efficiency, such as using model quantization and pruning to reduce inference energy consumption, as well as applying biomedical image compression to lower storage and bandwidth costs. Recently, I have also been exploring research in foundational models for omics data, aiming to bridge different modalities in biomedical research.

Why did you decide to participate in the France-BioImaging challenge “Light My Cells”?

I learned about the Challenge “Light My Cells” through my supervisor, who discovered the competition on X and shared it with me. I had never participated in a competition on the Grand Challenge platform before, so I wanted to experience the full process of such a challenge. Also we found the problem itself very interesting, and thanks to a previous project, we were already somewhat familiar with this topic.

What was the most challenging part of the competition for you?

The most challenging part of the competition was data processing. We used a semi-automated approach that combined an automated pipeline with manual curation for data filtering. This process was quite time-consuming.

How did you manage your time during the competition?

For the experimental phase, our team worked in parallel on several tasks: data cleaning, trying different network architectures, and conducting hyperparameter searches. However, when it came to writing the final paper, we were somewhat rushed as we had only about two weeks left to complete it.

What are your thoughts about Challenge 2 “Fuse My Cells”?

What excites me most about this challenge is its innovation. Predicting a fused 3D image directly from a single view bypasses some potential issues of multi-view fusion, such as light toxicity. Additionally, with a well-constructed dataset, the solution could be more generalizable, enabling models to perform image restoration across all the various 3D perspectives.

Do you have any advice for the participants of Challenge 2?

1. Pay attention to the data.
2. Keep an open mind and be willing to explore different strategies, including model selection and training methods.
3. Manage your time wisely by balancing the experimental phase and paper writing.

Thank you very much for your time, Yu! I’m sure your testimony will be useful for the participants of Challenge 2.

For those interested in taking part in the “Fuse My Cells” challenge, find more information here!