search

kdmalc / fl-gestures

Develop a personalized sequential federated learning algorithm for real-time gesture classification
1 stars 0 forks source link

Begin modeling latent gesture classification #19

Closed kdmalc closed 4 weeks ago

kdmalc commented 1 month ago
  1. Assign the nearest neighbor (KNN, k=1)
  2. KNN but k!=1
  3. Are either the gestures or users clustered in the latent space? If not, cluster-based approaches aren't going to do well...
  4. Traditional ML: SVM/RF
  5. Try a NN classification model: RNN/LSTM/FF
kdmalc commented 4 weeks ago

This was started in commit c4c9f9a92a8113361f3aec47e09dd26dabf892e4.

All the models perform quite poorly... (random chance is 3.57%), naive LSTM plateaus at 14%... need to do hyperparam tuning tho...