Autodidactic Neurosurgeon Collaborative Deep Inference for Mobile Edge Intelligence via Online Learning
Autodidactic Neurosurgeon (ANS) is an online learning module to automatically learn the optimal DNN partition point on the fly. The details of ANS are in our WWW21 paper "Autodidactic Neurosurgeon Collaborative Deep Inference for Mobile Edge Intelligence via Online Learning".
We modify the forward function in the PyTorch to partition the DNN model.
You can run "vgg16.py" as an example to see the partition.
Although we don't provide the code for TensorFlow, you can modify the _call_ function in the TensorFlow to partition the DNN model.
- vgg16
- tiny yolo v2
- Download tiny yolo weight, here https://pjreddie.com/media/files/yolov2-tiny-voc.weights
- Put yolov2-tiny-voc.weights in the folder "models"
- First run edge_server_main.py on edge server
- Then run client_camera_main.py on Nvidia Jetson TX2