Methods › General › Regularization › GradDrop

Gradient Sign Dropout

GradDrop

3 papers tagged archive 2025-07-28

Introduced by Zhao Chen et al. in Just Pick a Sign: Optimizing Deep Multitask Models with Gradient Sign Dropout

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

GradDrop, or Gradient Sign Dropout, is a probabilistic masking procedure which samples gradients at an activation layer based on their level of consistency. It is applied as a layer in any standard network forward pass, usually on the final layer before the prediction head to save on compute overhead and maximize benefits during backpropagation. Below, we develop the GradDrop formalism. Throughout, o denotes elementwise multiplication after any necessary tiling operations (if any) are completed. To implement GradDrop, we first define the Gradient Positive Sign Purity, 𝒫, as

𝒫=1/2(1+(∑ᵢ ∇L_{i)/∑ᵢ|∇Lᵢ|)

𝒫 is bounded by [0,1] . For multiple gradient values ∇ₐ Lᵢ at some scalar a, we see that 𝒫=0 if ∇ₐ Lᵢ<0 ∀i, while 𝒫=1 if ∇ₐ Lᵢ>0 ∀i. Thus, 𝒫 is a measure of how many positive gradients are present at any given value. We then form a mask for each gradient ℳᵢ as follows:

ℳᵢ=ℐ[f(𝒫)>U] ∘ℐ[∇Lᵢ>0]+ℐ[f(𝒫)<U] ∘ℐ[∇Lᵢ<0]

for ℐ the standard indicator function and f some monotonically increasing function (often just the identity) that maps [0,1] ↦[0,1] and is odd around (0.5,0.5). U is a tensor composed of i.i.d U(0,1) random variables. The ℳᵢ is then used to produce a final gradient ∑ℳᵢ ∇Lᵢ

PaperSource

Papers archive 2025-07-28

3 shown of 3, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

2 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Transfer Learning3
XLM-R1

Usage over time archive 2025-07-28

Papers per year tagged with GradDrop: 2020 to 2023, peak 1 1 0 2020: 1 paper 2020 2021: 1 paper 2021 2022: 0 papers 2022 2023: 1 paper 2023
Papers per year the archive tags with this method, by the paper's archive date (3 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Regularization

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections