On device training with TensorFlow Lite (Micro)

General Discussion
#1

I would like to understand the concept and the use of “on-device training” for new data acquired from sensor, camera, etc.
How to prepare new (unseen) data in terms of proper labeling? I assume that these data are unknown or unseen.
Do I need to prepare a batch of data, or can I just use a single sample?
Is this “On device training” applicable for supervised or unsupervised learning?
What are the limitations on On-Device Training with TensorFlow Lite
Is that concept somewhere described and explained (no reading or exploring code)?

Thank you for helping me to understand it!

#3

Hi @peter_cz,

  • Manually(User) can label the generalized data for training but while inferencing, the model will predict the label for unseen data.
  • It is recommended to prepare a batch of data for training a model as it allows the model to learn intricate and robust patterns.
  • On-devicetraining applicable for both supervised and unsupervised learning.
  • Inspite of the advantages of `On -device training’ : low latency, privacy and offline capability, the disadvantages are computational resource constraints means larger datasets training may not be feasible in terms of both memory and power. Here is the reference doc.

Thank You