Anomaly Detection Approach

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

Hi,

I have a signal comes in every ms in real time. And I have saved it in mySQL, (It looks like a heart beat waveform.) I wonder if there is any example of anomaly detection? Not just max or min, but the waveform like the duration of that pulse, and the shape of the pulse.

Shall I use LSTM like a speech? My dataset is only the normal one, I don’t have any abnormal dataset.
I have search the web and there is a list of tools. I wonder what is the right approach and has anyone implemented in the Nvidia nano.
https://github.com/Labaien96/Time-Series-Anomaly-Detection

Thanks,

3

Hi,

Sorry that we don’t have too much experiment on the anomaly detection.
Maybe you can check our healthcare page for some information:
https://developer.nvidia.com/clara-agx-devkit

Thanks.

5

Hi,

Thanks. I just want to know if jetson has or will have an anomaly detection (or LSTM) demo soon.
Or is there any jetson LSTM model that I can try? ('Coz I saw the Nvidia NX demo can recognize voice)

My friend and I have problem before porting the model from desktop to the nano. The fps drops a lot, is there any link that teach me how to convert a tensorflow model to tensorRT?

BTW, real-time detection will be great.

Thanks,

6

Hi,

We do have several built-in LSTM sample:

/usr/src/tensorrt/samples/sampleNMT
/usr/src/tensorrt/samples/sampleCharRNN

In sampleNMT, it uses the TensorFlow seq2seq model which may meet your requirement.

Thanks.