diff options
author | Nicholas <nbnoll@eml.cc> | 2022-02-24 15:57:52 -0800 |
---|---|---|
committer | Nicholas <nbnoll@eml.cc> | 2022-02-24 15:57:52 -0800 |
commit | 7f18d975a1b0ab091746021949520e3e294a1e77 (patch) | |
tree | 3f131eae51a0872edad979f6e6d860205500bc83 | |
parent | dcaa6396cf980f84356414ab71ed10590fd6f24a (diff) |
chore: update skeleton of autoencoder.md
-rw-r--r-- | docs/src/sci/autoencode.md | 10 |
1 files changed, 8 insertions, 2 deletions
diff --git a/docs/src/sci/autoencode.md b/docs/src/sci/autoencode.md index e8b4362..6a1e2ae 100644 --- a/docs/src/sci/autoencode.md +++ b/docs/src/sci/autoencode.md @@ -1,6 +1,12 @@ # Manifold learning -## Introduction +## Empirical analysis + +* Point cloud scaling +* Physically close cells are close in expression +* Isomap analysis picks up spatial gradients + +## Initial considerations Want: * Dimensional reduction @@ -13,7 +19,7 @@ Natural choice for an autoencoder. Utilize known positional labels as a validation step. Not used for training purposes. -## Architecture +## Network architecture How to pick depth? Width? Worry about overfitting: enter dropout and batch normalization. |