Learning-based Non-linear Erasure Code for Distributed Inference

erasure code.pdf: 项目的总体方向

FusedLayer_test.py: 它对应于pdf中给出的第一步操作, 给定output的range, 计算input的range

其他可能需要的代码:

• Learning-Based_Coded_Computation 的仓库:

  它对应于编码和解码的过程

  (1) https://github.com/Thesys-lab/learned-cc

  (2) https://github.com/Thesyslab/parity-models

• EdgeLD 的仓库:

  它对应于更后续的工作, 研究异质设备下, workload分配的部分

  https://github.com/fangvv/EdgeLD

Dataset and Base Model

• MNIST
  • LeNet-5: Gradient-based learning applied to document recognition
• CIFAR-10 / CIFAR-100
  • ResNet: Deep Residual Learning for Image Recognition
  • VGGNet: Very Deep Convolutional Networks for Large-Scale Image Recognition
  • DenseNet: Densely Connected Convolutional Networks
  • EfficientNet: EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
  • Wide ResNet: Wide Residual Networks

Encoder and Decoder Architecture

Encoder

input shape: $(k, a, a')$, output shape: $(r, a, a')$

• MLP with two layers: $k a a' \times k a a'$, $kaa' \times raa'$
• CNN with ? layers

Decoder

input shape: $(n, b, b')$, output shape: $(k, b, b')$

• MLP with three layers: $nbb' \times kbb'$, $kbb' \times kbb'$, $kbb' \times kbb'$
• CNN with ? layers