🗓️ Meeting Notes – 3 June 2025, 14:00

👥 Attendees

• David
• Pieter
• Zhendong
• Jules

---

🗒️ Last Meeting’s Feedback

David’s Updates

• David’s been working on hyperparameter tuning for both the graph augmentation and the GCN and GAT models.
• Results looked good; Zhendong recommended a graph to be created for next week.

Pieter’s Updates

• First runs with Hierarchical(multi-resolution) model. Added results to scoreboard.
• Went back to look at coarsening method. New implementation: Small connected components do not get a supernode.

General Discussion

• Try to use a density plot instead of/in addition to the scatter plot for results and comparison to the Palmes tubes.
• Provide the final baseline results after Optuna hyperparameter tuning.
• List all hyperparameters tuned and their final values.

---

🗒️ Progress

David’s Updates

• Ran Optuna-based hyperparameter tuning on both GCN and GAT models for 200 trials; added the new scores to the scoreboard.
• Attempted combined hyperparameter tuning (graph augmentation + model) for 400 trials on both models; results did not improve.
• Added clear documentation to all code and uploaded it to GitHub.
• Created the recommended plot from Zhendong.
• Obtained and finalized all results and code in order to begin writing the thesis.

---

🗒️ New Meeting Notes

Pieter’s Updates

• ...

General Discussion

• adamoptimizer maybe has 0.010 already implemented? Check how this works.
• disconnected sets of segments can cause noise, gps inaccuracy can be the cause of the 'disconnectedness'

---

📅 Next Steps

• David to …
• Pieter to …
• Zhendong to …
• Group to …

Next Meeting: 10 June 2025, ...