[src,scripts,egs] Add form of dropout that shares the mask across frames.#2244
Merged
danpovey merged 16 commits intokaldi-asr:masterfrom Mar 1, 2018
Merged
[src,scripts,egs] Add form of dropout that shares the mask across frames.#2244danpovey merged 16 commits intokaldi-asr:masterfrom
danpovey merged 16 commits intokaldi-asr:masterfrom
Conversation
Conflicts: egs/wsj/s5/steps/libs/nnet3/xconfig/basic_layers.py
Conflicts: src/nnet3/nnet-computation.cc
Contributor
Author
|
Just a note: For now this type of dropout seems to mostly be useful for very small-data setups, like mini-librispeech. @freewym, I have included a "block-dim" configuration value in this component that should be useful in applying it to image recognition setups (where it should be set to the num-filters)... this would be basically like GroupOut. It would be good if you could test that. |
Contributor
|
OK, will test it.
…On Wed, Feb 28, 2018 at 7:45 PM, Daniel Povey ***@***.***> wrote:
Just a note:
I was hoping by now to also have recipes with dropout for WSJ and
Switchboard. I thought I was getting improvements from it on those setups,
as well as mini_librispeech, but experiments @GaofengCheng
<https://github.com/gaofengcheng> has done seem to show that most or all
of that improvement was due to other tuning changes.
We may update those recipes soon though.
For now this type of dropout seems to mostly be useful for very small-data
setups, like mini-librispeech. @freewym <https://github.com/freewym>, I
have included a "block-dim" configuration value in this component that
should be useful in applying it to image recognition setups (where it
should be set to the num-filters)... this would be basically like GroupOut.
It would be good if you could test that.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#2244 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/ADWAknRe7qEXnjM2g8xw6yS7MgeCWZ5Xks5tZ0ShgaJpZM4SXk8P>
.
--
Yiming Wang
Department of Computer Science
The Johns Hopkins University
3400 N. Charles St.
Baltimore, MD 21218
|
LvHang
pushed a commit
to LvHang/kaldi
that referenced
this pull request
Apr 14, 2018
…mes. (kaldi-asr#2244) Conflicts: egs/mini_librispeech/s5/local/chain/run_tdnn.sh egs/wsj/s5/local/chain/tuning/run_tdnn_1f.sh egs/wsj/s5/steps/nnet3/chain/train.py src/chain/chain-training.cc src/nnet3/nnet-chain-training.cc src/nnet3/nnet-compute.cc src/nnet3/nnet-normalize-component.h src/nnet3/nnet-training.cc
Skaiste
pushed a commit
to Skaiste/idlak
that referenced
this pull request
Sep 26, 2018
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.