Discussion: state-of-the-art ML methods

[YoniSchirris]

• Squeezenet: A lightweight image classifier
• Resnet for time series analysis of PPEG

[LittlePea13]

Working on CNNs branch. Currently on Semcova dataset: Lowest MAE for Spo2 on test set, paper reported 1.1%:

• Mobilenet_v2: 0.85%
• Resnet18: 1.05%
• VGG13: 0.95%
• VG16: 0.87%

Please be aware I was reporting the error at each epoch, and these are the best epoch results, before it starts over-fitting. Since we do not have a validation set (data to scarce), this is an overly optimistic results. There is also not enough training data to call this anywhere conclusive, but we have the models available to train once we get more data.

I have not performed training on HR, although the network output is both spo2 and HR, HR error was quite big compared to spo2 when I first started testing, so for now I only train on spo2 (loss signal is restricted to it), but all nets should be able to be trained simultaneously for both values.