23 lines
1.2 KiB
Markdown
23 lines
1.2 KiB
Markdown
This is the coding part of the lecture Deep Representation Learning in PyTorch.
|
||
|
||
python 3.10
|
||
|
||
#### Autoencoder
|
||
In this demo we will implement a simple autoencoder. The autoencoder will be trained on the MNIST dataset. The autoencoder will be implemented in the file autoencoder.py. The file autoencoder.py contains a class Autoencoder on the MNIST dataset.
|
||
We compare the performance of the fully connected autoencoder with a convolutional autoencoder.
|
||
Jupyter notebooks:
|
||
* autoencoder_mnist.ipynb
|
||
* autoencoder_cifar10.ipynb
|
||
|
||
#### Contractive Learning (SimCLR)
|
||
In this demo we implemented the SimClR [1] algorithm and trained it on the cifar10 dataset.
|
||
|
||
Download the pretrained encoder [here](https://cloud.cps.unileoben.ac.at/index.php/s/feHYqRHwDy7mMDm) and put it in the folder `runs`.
|
||
|
||
The code was adapted from this [repo](https://github.com/sthalles/SimCLR/tree/master)
|
||
|
||
Jupyter notebooks:
|
||
* simclr.ipynb
|
||
|
||
|
||
[1] Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In Proceedings of the 37th International Conference on Machine Learning (ICML'20), Vol. 119. JMLR.org, Article 149, 1597–1607. |