tensorflow / tensorflow/probability
Regularization in FFJORD
Nobody has claimed this yet.
- Dominant language
- Jupyter Notebook
- Stars
- 4.4k
- Forks
- 1.1k
- PR merge metrics
- No merged PRs in 30d
Description
It would be great to implement regularization techniques that have been recently been developed for FFJORD. For example,
The simplest regularization adds a penalty for the path length of each sample's trajectory. I don't know how to begin to implement this in Tensorflow Probability, as I don't understand the internals well enough. During training time, when calculating log_prob(batch), it's necessary to calculate the integral of |f(t,x)| over the path taken by each sample in the batch, and to somehow expose this information so that it can be used in an additional penalty term.
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 reading the linked regularization paper and tracing the TensorFlow Probability FFJORD log_prob(batch) path described in the issue. Done means the path-length penalty can be calculated for each sample during training and the required information is exposed for an additional penalty term.
Written by the indexing model from the issue text.
Assessment
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100