Comments (3)
So turns out this also makes a difference between training-time and testing-time, just like Dropout #2 . Guess I'll have to refactor a bit so I can easily implement both of them.
from deepfried.
The refactor is happening simultaneously to the export =)
from deepfried.
Turns out this is harder than that, and not "just like Dropout." Ideally, it'd work in the following way:
- Train the network, nothing special
- Build the prediction network by:
- Sending all training data through the trained network up to the first BN-layer.
- Compute the training's mean and variance here.
- Use these to compute this BN's prediction γ and β as in Line 11 of the paper.
- Sending all training data through the trained network up to the second BN-layer.
- ...
As an easy stop-gap, one can use the prediction's minibatch statistics. People report that working well, it just makes prediction nondeterministic.
from deepfried.
Related Issues (18)
- Implement Dropout HOT 4
- Use multiprocessing for augmentation
- Alex' PCA-RGB augmentation HOT 1
- Zooming augmentation HOT 1
- Precompile prediction augmentations
- Implement 'same' convolution HOT 1
- Implement 1x1 convolutions The Right Way™
- Make Containers like in Torch
- Implement alternatives to SGD HOT 2
- Port all layers from lbtoolbox
- Implement L1- and L2-weight penalties.
- Port the trainer from lbtoolbox HOT 1
- Write documentation HOT 1
- Port the reinitialization behaviour from lbtoolbox
- Add a verbose switch? HOT 1
- Augmenters should take `rng` param.
- Shuffling
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from deepfried.