search

Edge-Learning-Machine / Autonomous-Edge-Pipeline

AEP is a self-learning autonomous edge learning and inferencing pipeline for resource-constrained embedded system.
1 stars 0 forks source link

Both unsupervised learning and supervised learning are on the device side #1

Open elimsjxr opened 2 years ago

elimsjxr commented 2 years ago

Is it correct for me to understand this way? You are implementing kmeans on stm32, through kmeans, the data received by the sensor is divided into several categories and labeled, and after a certain amount is collected, it will be trained on the device side, using KNN？Both unsupervised learning and supervised learning are on the device side?

FouadSakr commented 2 years ago

Yes and yes. Also, the clustering and classification are binary.

In general, the AEP is an iterative pipeline for resource-constrained devices that alternates between clustering, training, and classification. The k-means is periodically executed on the input stream and provides the pseudo-labeled clustered data. The labeling results are then evaluated by a confidence algorithm (if enabled by the user), which makes a binary decision whether to keep or discard each instance. Once important samples with their corresponding pseudo-labels are selected, the training process is executed (this applies to the DT case only, since K-NN does not involve a training phase). The resulting classifier then continuously classifies the incoming samples. To prevent device memory overflow, we have also implemented some memory management strategies.

If anything in the readme file is unclear, please let us know. Your suggestions for improvement are also welcome. We will also post the paper as soon as it is published.