LayerAct: Advancing CNNs with BatchNorm through Layer-direction Normalization

Kihyuk Yoon, Chiehyeon Lim

Primary Area: general machine learning (i.e., none of the above)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: deep learning, activation function, cnn, convolutional neural network, batch normalization, layer-direction normalization, noise-robustness

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: This paper introduces a novel activation mechanism and two LayerAct functions for CNNs with BatchNorm to address two limitations of existing activation functions: i) trade-off problem, and ii) large variance of noise-robustness across samples.

Abstract: In this work, we propose a novel activation mechanism aimed at establishing layer-level activation (LayerAct) functions for CNNs with BatchNorm. These functions are designed to be more noise-robust compared to existing element-level activation functions by reducing the layer-level fluctuation of the activation outputs due to shift in inputs. Moreover, the LayerAct functions achieve this noise-robustness independent of the activation's saturation state, which limits the activation output space and complicates efficient training. We present an analysis and experiments demonstrating that LayerAct functions exhibit superior noise-robustness compared to element-level activation functions, and empirically show that these functions have a zero-like mean activation. Experimental results with three benchmark datasets for image classification tasks show that LayerAct functions excel in handling noisy datasets, outperforming element-level activation functions, while the performance on clean datasets is also superior in most cases.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7871