Adv training #24
Merged
Adv training #24
Conversation
This file contains 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
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #24 +/- ##
==========================================
- Coverage 41.32% 40.66% -0.66%
==========================================
Files 10 10
Lines 242 241 -1
==========================================
- Hits 100 98 -2
- Misses 142 143 +1 ☔ View full report in Codecov by Sentry. |
pat-alt
reviewed
Sep 5, 2024
| ) | ||
| θ = Flux.params(model) | ||
| vanilla_losses = [] | ||
| train_loader = DataLoader((x_train, y_train), batchsize = batch_size, shuffle = true) | ||
| train_loader = DataLoader((x_train, y_train), batchsize = batch_size, shuffle = true) |> gpu |
There was a problem hiding this comment.
Should this be Flux.get_device() to keep it general? https://fluxml.ai/Flux.jl/stable/guide/gpu/#Using-device-objects
There was a problem hiding this comment.
Neat, didn't know that was a thing. I'll change this, and I'll also add some tests for the train, and the attacks (simple ones for now to check if the perturbation is within bounds)
| clamp_range = (0, 1), | ||
| ) | ||
| adv_losses = [] | ||
| θ = Flux.params(model) | ||
| train_loader = DataLoader((x_train, y_train), batchsize = batch_size, shuffle = true) | ||
| train_loader = DataLoader((x_train, y_train), batchsize = batch_size, shuffle = true) |> gpu |
pat-alt
approved these changes
Sep 10, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Solves issue #23. I have described as well as I can the issue, let me know if it's confusing.
Currently seems as though FGSM training actually works. Will test PGD training soon.
Edit: PGD training seems to work too. Only thing is if the adversary in train-time is too strong, the model doesn't seem to want to learn anything.
Also based on a few runs, what i notice looks like it makes sense. Classical training performs poorly on both FGSM and PGD adversaries. FGSM training does decent against FGSM adversary but poorly against PGD. PGD training did well against both. Checks out