July 28, 2025
Day 45 - Refining Results and Image Selection for Presentation
What I Learned
Today’s internship session was focused on tightening the final stages of our ECG project. I dedicated a significant portion of the day to updating the Results section of our work, making sure that all key performance metrics — including accuracy, precision, recall, and confusion matrices — were properly formatted and clearly reflected the model performance across the different modalities (1D CNN, 2D CNN, and hybrid fusion approaches). I made sure to adjust any outdated figures and correct previous inconsistencies so that our results now accurately represent the models we trained and validated. A major focus was also on selecting the right images for our presentation and report. This meant going through the folders of visual outputs — including spectrograms, Grad-CAM maps, and confusion matrices — and carefully curating the ones that best illustrated our findings. I prioritized clarity, uniqueness, and relevance to the core claims we’re making. The goal was to ensure that each image adds visual strength and supports the narrative we’re presenting, especially for explaining model interpretability and class performance. Overall, today was about preparing our project to be publication- and presentation-ready — refining the results and visuals so they clearly reflect the value and depth of our work. The process involved both technical updates and thoughtful curation, and it helped bring us closer to the final version of our report.
Blockers
No blockers.
Reflection
Today’s work reminded me that sometimes the hardest part isn’t building the models — it’s communicating what we’ve built clearly. I spent most of the day diving into the Results section of our project. It felt like cleaning up after a long experiment — making sure every metric we reported actually reflected the most recent version of our models. I updated the accuracy scores, double-checked the confusion matrices, and made sure the layout was clean and easy to follow. It was tedious, but necessary — I realized that this is what turns a good project into a presentable one. One of the more surprisingly challenging parts was choosing the right images. We had so many — spectrograms, CAM heatmaps, denoising outputs — and I had to go through them one by one, deciding which ones told the best story. It wasn’t just about picking the clearest images — it was about picking the right ones to communicate key ideas visually. I found myself asking: If someone knew nothing about this project, would this image help them understand? In a way, today was about stepping back and looking at the bigger picture. It wasn’t about code or models — it was about narrative. I’m learning that in research, clarity is just as important as complexity. You can have the best model in the world, but if no one understands what you did or why it matters, it’s as if it never existed.