AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Excitation Dropout: Encouraging Plasticity in Deep Neural Networks

Andrea Zunino, Sarah Adel Bargal, Pietro Morerio, Jianming Zhang, Stan Sclaroff, Vittorio Murino

IJCV 2021 · Architectures and efficiency › RNNs and normalization · approach: Representation geometry and collapse · 25 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add excitation-based retaining probabilities (vs standard dropout, both fcs)Dropout in fc layers · UCF101 AlexNettest accuracy (%): 65.83 -> 67.56Improved
use input-specific masks guided by excitation rather than random masks per image (ablation)Standard Dropout · Cifar10/Cifar100/Caltech256test accuracyImproved
apply Excitation Dropout at conv3 instead of fc1 (ablation)Excitation Dropout · Cifar10 (CNN-2)test accuracy (%): 81.94 -> 78.01Hurt
apply Excitation Dropout at conv3 instead of no dropout (ablation)CNN-2 · Cifar10test accuracy (%): 76.91 -> 78.01Improved
replace Excitation Dropout by Activation Dropout (ablation)Excitation Dropout · Cifar10test accuracy (%): 81.94 -> 81.33Hurt
replace Excitation Dropout by Activation Dropout (ablation)Excitation Dropout · Caltech256test accuracy (%): 35.77 -> 35.29Hurt
replace Excitation Dropout by Activation Dropout (ablation)Excitation Dropout · UCF101 VGG16test accuracy (%): 73.23 -> 72.56Hurt
use Activation Dropout instead of Standard Dropout on WRN-28-10WideResNet-28-10 · Cifar10test error (%): 4.17 -> 3.88Improved
use Activation Dropout instead of Standard Dropout on AlexNetAlexNet from scratch · ImageNettest accuracy (%): 44.43 -> 48.54Improved
add Excitation Dropout instead of Standard Dropout in training timeStandard Dropout · VGG16 trainingtime per 100 iterations (s): 2.279 -> 2.820Hurt
increase dropout rate from 0.25 to 0.75 for Excitation Dropout (ablation)Excitation Dropout · Cifar10test accuracy (%): 81.38 -> 81.55Mixed
increase dropout rate from 0.25 to 0.75 for Standard Dropout (both fcs) (ablation)Standard Dropout · Cifar100test accuracy (%): 50.30 -> 49.46Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy