3 layers cloud computing, First, we use vargrant to enable three virtual machines as front-end, back-end, and data base.
- try classfication model with
CIFAR10
- Learned
- write dataset class, dataloader
- data transform with pytorch
- build CNN model
- simple train, test part
- practice classfication model with
Mnist, KMnist, Flower102
- learned
- dynamic add transform when training phase
- use
PrintLayer
to debug model layer - Comparing the effects of different activation finction
relu, leakyRelu, elu
- Accelerating training with
amp.GradScaler()
- Implement FN from scratch
- learned
- how NN work (forward, backward)
- Solving simple NLP Problems with LSTM
- learned
- how LSTM worked (
EOS, SOS, inference
) embedding layer, position encoding
- Impact of
teacher forcing
- how LSTM worked (
- preporcess binary dataset
EMnist
, and try 3 generative modelDCGAN, Cycle GAN, Conditional diffusion model
- learned
- preporcess binary dataset
EMnist
to npy file - implement
residual Block
- implement
resNet
- Compare cycle gan different training methods
- try
Conditional diffusion model
- preporcess binary dataset
- implment
Unet
from scratch, and try to segment, classifyCCAgT
dataset - learned
- design an
Unet model
- data preprocess
process two img with same transform
,find interested part and crop
- implement
attention Unet model
,attention gate
- use
dice loss
- design an