tensorflow / tensorflow/probability
Get decent nats and generation in VAE example
Open
Nobody has claimed this yet.
- Dominant language
- Jupyter Notebook
- Stars
- 4.4k
- Forks
- 1.1k
- PR merge metrics
- No merged PRs in 30d
Description
The current VAE example is minimally done to show that it "runs". We should significantly improve it in multiple ways if we think this is the official approach people should take to implement VAEs in TF:
- Dataset. Binarized MNIST is a good first step. We should at least have competitive numbers on CIFAR-10 too.
- Architectures. Standard normal prior, 2-layer feedforward, and conditionally independent likelihood isn't very good. We can extend all three.
- Hyperparameters. Latent size, hidden units, batch size, and many more.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by locating the current VAE example and reviewing how it handles the dataset, architecture, and hyperparameters. Define the intended approach before changing it; done should include binarized MNIST, competitive CIFAR-10 results, stronger prior and architecture choices, and documented hyperparameter improvements.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- jupyter-notebook
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100