Papers › Trainable Activations for Image Classification

Trainable Activations for Image Classification

26 Jan 2023Preprints 2023 1archive 2025-07-28

Evgenii Pishchik

Non-linear activation functions are one of the main parts of deep neural network architectures. The choice of the activation function can affect model speed, performance and convergence. Most popular activation functions don't have any trainable parameters and don't alter during the training. We propose different activation functions with and without trainable parameters. Said activation functions have a number of advantages and disadvantages. We'll be testing the performance of said activation functions and comparing the results with widely known activation function ReLU. We assume that the activation functions with trainable parameters can outperform functions without ones, because the trainable parameters allow the model to "select'' the type of each of the activation functions itself, however, this strongly depends on the architecture of the deep neural network and the activation function itself.

PaperPDFCode

Code

Pe4enIks/TrainableActivation officialmentioned in paperpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Image Classificationimage-classification

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Classification CIFAR-10 ResNet-26 (Trainable Activations) Percentage correct 91.1 #196 of 265 Archive leaderboard report
Image Classification CIFAR-10 ResNet-32 (Trainable Activations) Percentage correct 90.9 #197 of 265 Archive leaderboard report
Image Classification CIFAR-10 ResNet-44 (Trainable Activations) Percentage correct 90.5 #204 of 265 Archive leaderboard report
Image Classification CIFAR-10 ResNet-20 (Trainable Activations) Percentage correct 90.4 #205 of 265 Archive leaderboard report
Image Classification CIFAR-10 ResNet-14 (Trainable Activations) Percentage correct 89.0 #212 of 265 Archive leaderboard report
Image Classification CIFAR-10 ResNet-56 (Trainable Activations) Percentage correct 88.8 #215 of 265 Archive leaderboard report
Image Classification CIFAR-10 ResNet-8 (Trainable Activations) Percentage correct 86.5 #228 of 265 Archive leaderboard report
Image Classification MNIST DNN-5 (Trainable Activations) Accuracy 97.2 #59 of 81 Archive leaderboard report
Image Classification MNIST DNN-5 (Trainable Activations) Percentage error 2.8 #59 of 81 Archive leaderboard report
Image Classification MNIST DNN-5 (Trainable Activations) Trainable Parameters 575051 #59 of 81 Archive leaderboard report
Image Classification MNIST DNN-3 (Trainable Activations) Accuracy 97.0 #60 of 81 Archive leaderboard report
Image Classification MNIST DNN-3 (Trainable Activations) Percentage error 3.0 #60 of 81 Archive leaderboard report
Image Classification MNIST DNN-3 (Trainable Activations) Trainable Parameters 386719 #60 of 81 Archive leaderboard report
Image Classification MNIST DNN-2 (Trainable Activations) Accuracy 96.4 #61 of 81 Archive leaderboard report
Image Classification MNIST DNN-2 (Trainable Activations) Percentage error 3.6 #61 of 81 Archive leaderboard report
Image Classification MNIST DNN-2 (Trainable Activations) Trainable Parameters 311651 #61 of 81 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: CosLU, DELU, LinComb, NormLinComb, ReLUN, ScaledSoftSign, ShiLU

1x1 ConvolutionAverage PoolingBatch NormalizationBottleneck Residual BlockConvolutionCosLUDELUGlobal Average PoolingKaiming InitializationLinCombMax PoolingNormLinCombReLUReLU6ReLUNResidual BlockResidual ConnectionSGDScaledSoftSignShiLUSiLUSigmoid ActivationSoftsign ActivationTanh Activation

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