T$_k$CP: Context-Aware Pooling via Top-k% Activation Selection

Seo-Yeon Choi, Kyungsu Lee
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:3502-3510, 2026.

Abstract

Pooling is a core operation in convolutional neural networks (CNNs), enabling spatial reduction and hierarchical abstraction. However, standard methods such as max or average pooling operate locally and often fail to capture global context, leading to under- or over-estimation of features. This limits performance on tasks requiring both fine localization and holistic understanding. To address this, we propose Top-$k$% Contextual Pooling (TkCP), a framework that preserves informative activations based on contextual importance. TkCP includes two variants: (1) Sparse Contextual Pooling, selecting top-$k$% activations within local windows, and (2) Global Contextual Pooling, selecting top-$k$% across the entire feature map. Given a kernel size and target resolution, TkCP deterministically sets the stride and reconstructs outputs without additional parameters. Experiments across classification, detection, tracking, segmentation, and generation show consistent improvements in accuracy and robustness. Additionally, TkCP enhances interpretability by tracing high-activation regions across layers.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-choi26b, title = { T$_k$CP: Context-Aware Pooling via Top-k% Activation Selection }, author = {Choi, Seo-Yeon and Lee, Kyungsu}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {3502--3510}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/choi26b/choi26b.pdf}, url = {https://proceedings.mlr.press/v300/choi26b.html}, abstract = { Pooling is a core operation in convolutional neural networks (CNNs), enabling spatial reduction and hierarchical abstraction. However, standard methods such as max or average pooling operate locally and often fail to capture global context, leading to under- or over-estimation of features. This limits performance on tasks requiring both fine localization and holistic understanding. To address this, we propose Top-$k$% Contextual Pooling (TkCP), a framework that preserves informative activations based on contextual importance. TkCP includes two variants: (1) Sparse Contextual Pooling, selecting top-$k$% activations within local windows, and (2) Global Contextual Pooling, selecting top-$k$% across the entire feature map. Given a kernel size and target resolution, TkCP deterministically sets the stride and reconstructs outputs without additional parameters. Experiments across classification, detection, tracking, segmentation, and generation show consistent improvements in accuracy and robustness. Additionally, TkCP enhances interpretability by tracing high-activation regions across layers. } }
Endnote
%0 Conference Paper %T T$_k$CP: Context-Aware Pooling via Top-k% Activation Selection %A Seo-Yeon Choi %A Kyungsu Lee %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-choi26b %I PMLR %P 3502--3510 %U https://proceedings.mlr.press/v300/choi26b.html %V 300 %X Pooling is a core operation in convolutional neural networks (CNNs), enabling spatial reduction and hierarchical abstraction. However, standard methods such as max or average pooling operate locally and often fail to capture global context, leading to under- or over-estimation of features. This limits performance on tasks requiring both fine localization and holistic understanding. To address this, we propose Top-$k$% Contextual Pooling (TkCP), a framework that preserves informative activations based on contextual importance. TkCP includes two variants: (1) Sparse Contextual Pooling, selecting top-$k$% activations within local windows, and (2) Global Contextual Pooling, selecting top-$k$% across the entire feature map. Given a kernel size and target resolution, TkCP deterministically sets the stride and reconstructs outputs without additional parameters. Experiments across classification, detection, tracking, segmentation, and generation show consistent improvements in accuracy and robustness. Additionally, TkCP enhances interpretability by tracing high-activation regions across layers.
APA
Choi, S. & Lee, K.. (2026). T$_k$CP: Context-Aware Pooling via Top-k% Activation Selection . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:3502-3510 Available from https://proceedings.mlr.press/v300/choi26b.html.

Related Material