June 23, 2025
Day 20 – Monday - not so 'fun'day
What I Learned
Today was just another laid-back day. Our grad mentor Pelumi is in Canada for a conference, so our previously assigned mentor, Ms. Oyinkansola Aladeokin, will be supervising us for the rest of the week. I jumped onto my PC as soon as I arrived. I had trained a model last Friday and left the PC on, but it turns out the electricity in the CBEIS building was cut off for scheduled maintenance, interrupting my model training. Unfortunately, Jupyter didn’t autosave the progress, so I had to restart the training. While randomly checking our image dataset, I discovered some anomalies. I found images in the non-drowsy set that resemble the drowsy state. The examples shown in the attachment are just a few observations; there might be many more misleading images in the wrong set.
This issue is concerning and will definitely affect the credibility of our work.
Blockers
The PC abruptly shut down on Saturday due to scheduled electrical maintenance, resulting in the loss of all my training progress.
Reflection
Today was a very hot day, and for the first 2 hours, it was unbearable inside the lab, especially with the AC not working properly. I lost my previous training progress, which has left me quite frustrated as I had to restart the training, a process that will take the whole day. On top of this, I identified some major flaws in our dataset, that has me question the credibility of our model and overall research, as training the model on an incorrect dataset will inevitably yield inaccurate or poor results.