The issue of online learning resource optimization for models generated by NanoEdge AI Studio in sound anomaly detection.

[ScottO]

Hello!

    I plan to use NanoEdge AI Studio for sound anomaly detection, but currently I have some confusion: the resources of embedded chips are limited, and the amount of sound signal sampling data is relatively large. Can online learning be implemented on hardware. If feasible, the chip should not be able to learn much sample size. Can this ensure high accuracy.

    Can you help me solve the problem.Thank you.

[Julian E.]

Hello @ScottO,

I don't know how familiar you are with NanoEdge, but the way Anomaly detection project is made is as follow:

1. You determine a sampling frequency and a buffer size you will use in your project (generally it is an iterative process)
2. You use the NanoEdge Benchmark to find a library with hopefully a good accuracy
3. You do some test to make sure that the library is working on new data
4. You deploy the library
5. You can choose to retrain the library with nominal data

The first training done in NanoEdge, with your huge data base. It is used to determine the best kind of model, preprocessing and parameters. 

The way you retrain an embedded model is by giving new signal to learn. One signal at a time.

The retraining is very useful to adapt specifically to multiple machines. For example, if you have 10 motors of the same kind, the retraining will allow to fit perfectly to these 10 motors that will have slightly different vibrations pattern.

I hope it answers your question. 

Please let me know if you need further explanations.

Have a good day,

Julian