---
title: Hybrid Pooling Downsampling (HPD)
url: https://www.emergentmind.com/topics/hybrid-pooling-downsampling-hpd
type: topic
---

# Hybrid Pooling Downsampling (HPD)

Searching arXiv for the cited HPD-related papers to ground the article and citations.
Hybrid Pooling Downsampling (HPD) denotes a family of downsampling strategies that replace fixed, single-criterion reduction with hybrid operators intended to preserve more task-relevant information during resolution reduction. The term is not used canonically across the literature: it has been used for a hybrid of DiffStride and Spectral Pooling in convolutional networks [2401.09008], for graph-level readout operators that fuse attention, sum, and max pooling in mixup-based graph learning [2210.03123], for a MinMaxPooling-based encoder module in medical image segmentation [2507.14790], and more broadly for a design viewpoint in which downsampling mixes pooling criteria, multi-scale structure, or pre-filtering [2009.07485]. This suggests that HPD is best understood as an architectural principle—hybridization of downsampling mechanisms to improve information retention—rather than as a single fixed operator.

## 1. Terminological scope and conceptual basis

Pooling layers are described as the component that executes down-sampling on feature maps, drastically reduces spatial dimension, reduces the number of parameters or weights, lessens computational cost, and controls overfitting [2009.07485]. Within that general role, HPD reframes downsampling as a learnable or structured aggregation problem rather than a fixed application of max pooling, average pooling, or constant stride.

| Setting | HPD meaning | Representative source |
|---|---|---|
| CNN design viewpoint | mixing multiple pooling mechanisms, multi-scale priors, or pre-filtering | [2009.07485] |
| CNN image classification | combined implementation of Spectral Pooling and DiffStride on ResNet-18 | [2401.09008] |
| Graph learning | fusion of multiple graph readouts into a graph-level embedding | [2210.03123] |
| Semantic segmentation | MinMaxPooling with lightweight feature filtering | [2507.14790] |

In regular CNNs, the motivation is recurrent: fixed stride samples on a rigid grid, max pooling discards submaximal responses, and average pooling can blur edges or suppress salient activations [2401.09008], [2103.01746], [2009.07485]. In graph settings, the motivation is analogous but expressed at the level of node-set compression or topology-preserving coarsening, where the objective is to reduce graph size while preserving semantic information or structural properties [2210.03123], [2208.03523]. Across these usages, “hybrid” may mean mixing max and average behavior, combining learned stride with spectral truncation, fusing minimum and maximum local responses, or coupling structural coarsening with feature pooling.

## 2. Regular-domain formulations in convolutional models

A central CNN instantiation appears in "Hybrid of DiffStride and Spectral Pooling in Convolutional Neural Networks" [2401.09008]. The starting point is standard strided convolution,
$$
y[n] = \sum_k w[k]\,x[n\cdot s + k],
$$
where fixed $s$ can skip informative features. In that formulation, DiffStride learns stride parameters $S=(S_h,S_w)$ by backpropagation through a frequency-domain downsampling operator. The output size is coupled to the learned stride as
$$
H' = \lfloor H / S_h + 2R \rceil,\qquad W' = \lfloor W / S_w + 2R \rceil,
$$
with $R$ a smoothness factor. Spectral Pooling is defined by transforming to the frequency domain, cropping low frequencies to a target size, and inverting:
$$
y=T(x),\qquad \hat y = CropSpectrum(y; h\times w),\qquad \hat x=T^{-1}(\hat y).
$$
The hybrid places DiffStride in early and mid downsampling locations of ResNet-18 and inserts Spectral Pooling one level above global average pooling. The authors report that placing Spectral Pooling above GAP is more effective than placing a second DiffStride there [2401.09008].

A second convolutional formulation appears in "A Novel Downsampling Strategy Based on Information Complementarity for Medical Image Segmentation" [2507.14790]. Here HPD is a MinMaxPooling module that extracts complementary bright and dark information. For input $X\in\mathbb{R}^{C\times H\times W}$, the hard local statistics are
$$
y_{\max}(c,i,j)=\max_{(u,v)\in W(i,j)} X(c,u,v),\qquad
y_{\min}(c,i,j)=\min_{(u,v)\in W(i,j)} X(c,u,v),
$$
followed by either weighted fusion,
$$
y(c,i,j)=\alpha\,y_{\max}(c,i,j)+\beta\,y_{\min}(c,i,j),
$$
with $\alpha\ge 0$, $\beta\ge 0$, $\alpha+\beta=1$, or concatenation plus a $1\times 1$ convolution, batch normalization, and ReLU. The paper also gives optional soft min/max approximations through LogSumExp to alleviate gradient sparsity [2507.14790].

A third regular-domain formulation is provided by AdaPool, explicitly described as an HPD instantiation [2111.00772]. AdaPool fuses two exponential kernels per pooling window: an exponentiated Dice–Sørensen-coefficient weighting and an exponential maximum. Its regional fusion is
$$
\widetilde{\mathbf a}_{ada}(R)=
\widetilde{\mathbf a}_{eDSC}(R)\cdot \beta(R)+
\widetilde{\mathbf a}_{eM}(R)\cdot \big(1-\beta(R)\big),
$$
where $\beta(R)$ is a learned regional mask. The stated objective is to avoid average pooling’s oversmoothing and max pooling’s extreme selection while retaining non-zero gradients for all positions [2111.00772].

The broader review literature places these methods within a continuum of hybridization strategies. Mixed pooling, rank-based pooling, weighted pooling, adaptive $L_p$ pooling, anti-aliased blur pooling, spatial pyramid pooling, and spectral pooling are all treated as mechanisms that tune the information-loss versus invariance trade-off, often by interpolating between average-like and max-like behavior or by adding pre-filtering before decimation [2009.07485]. By contrast, "Comparison of Methods Generalizing Max- and Average-Pooling" restricts HPD to local window operators that interpolate between max and average, including Gated Pooling, Ordinal Pooling, Learned Norm Pooling, and Smooth-Maximum Pooling [2103.01746].

## 3. Graph-based formulations and structural generalization

In language-processing GNNs, HPD refers to hybrid graph pooling operators used as global readout rather than spatial reduction in a grid. "On the Effectiveness of Hybrid Pooling in Mixup-Based Graph Learning for Language Processing" defines node updates and graph readout as
$$
H_u^{(k)} = GNN\!\left(H_u^{(k-1)}, W, \sum_{v\in N(u)} H_v^{(k-1)}\right),\qquad
H^G = P(H^{(K)}, V).
$$
The study uses global readout without top-$k$ selection or soft cluster assignment. Its single-criterion readouts are attention pooling, sum pooling, and max pooling, and the hybrid operators fuse two readouts by elementwise sum, Hadamard product, or concatenation plus linear projection:
$$
M_{sum}(P_i,P_j)=H_{P_i}^G+H_{P_j}^G,\qquad
M_{mul}(P_i,P_j)=H_{P_i}^G\odot H_{P_j}^G,
$$
$$
M_{concat}(P_i,P_j)=T([H_{P_i}^G \parallel H_{P_j}^G])+b_c.
$$
Nine hybrid operators are evaluated. Type 1, $M_{sum}(P_{att},P_{max})$, and Type 3, $M_{concat}(P_{att},P_{max})$, are repeatedly identified as strong choices, while Type 6, $M_{concat}(P_{att},P_{sum})$, is singled out in some robustness settings [2210.03123].

A structurally different graph interpretation appears in "Generalizing Downsampling from Regular Data to Graphs" [2208.03523]. There HPD can be instantiated as topology-preserving graph coarsening paired with feature pooling. The coarsening selects centroids through a maximal $k$-independent set, induces clusters around them, contracts edges between clusters, and pools node features inside each cluster. The parameter $k$ acts as a stride or resolution control: centroids are at least $k+1$ hops apart, every node is within $k$ hops of a centroid, and larger $k$ yields coarser representations [2208.03523].

This graph-coarsening view is notable because it provides formal correspondence to regular downsampling on diagonal grid graphs. With lexicographic ranking, the paper states that $k$-MIS selects the same positions as strided sampling with stride $s=k+1$, and that the induced partition matches pooling windows of size $p=k+1$ [2208.03523]. In this usage, HPD is hybrid because it combines structural downsampling and feature pooling rather than because it fuses multiple scalar pooling operators.

## 4. Empirical performance across tasks

The empirical record for HPD is heterogeneous rather than uniformly positive. In the DiffStride–Spectral Pooling hybrid on ResNet-18, the reported mean accuracy on CIFAR-10 increases from 0.924 for the DiffStride baseline to 0.9334 for the hybrid, an improvement of 0.0094, and the CIFAR-100 mean accuracy increases from 0.706 to 0.7382, an improvement of 0.0322 [2401.09008]. The same study reports that a two-DiffStride combination reaches 0.9282 on CIFAR-10, below the hybrid with Spectral Pooling [2401.09008].

In graph learning for language and code, hybrid readouts are reported to outperform MAXPOOL and GMT in both accuracy and robustness. On NLP tasks, Type 1 improves accuracy versus GMT by up to 4.38% on Politifact, and on Gossipcop it achieves average 96.56% versus GMT 95.71%. On code datasets, Type 1 outperforms GMT by up to 2.36% accuracy. Robustness gains are larger in some perturbation settings: Type 6 surpasses GMT by up to 23.23% on GIN with Gossipcop-Profile, and Type 1 improves robustness by up to 10.23% on Python800. The paper further reports that Type 1 is significantly better than No Aug, MAXPOOL, and GMT across most models for accuracy, and in 10/15 tested cases for robustness under Wilcoxon signed-rank testing with Bonferroni correction [2210.03123].

In semantic segmentation, the MinMaxPooling formulation reports gains on ACDC and Synapse across several backbones, with average improvement of about 0.5% mDSC across architectures [2507.14790]. Illustrative cases include UNet on ACDC, 0.8993 to 0.9032; LinkNet on ACDC, 0.8690 to 0.8767; Fast-SCNN on ACDC, 0.7753 to 0.7863; and Swin-UNet on Synapse, 0.6227 to 0.6531. The paper also reports negative or marginal cases, including Deeplabv3+ on ACDC, 0.8001 to 0.7901, and Fast-SCNN on Synapse, 0.6487 to 0.6465 [2507.14790].

AdaPool reports systematic gains across multiple vision tasks. On ImageNet classification, the paper gives top-1 improvements such as ResNet-18 from 69.76 to 71.78, ResNet-50 from 76.15 to 78.42, DenseNet-121 from 74.65 to 77.29, ResNeXt-50 32×4d from 77.62 to 79.98, and Wide-ResNet-50 from 78.51 to 80.24. On MS COCO, Mask R-CNN with ResNet-101 is reported to improve AP from 38.2 to 42.4. The same work also reports gains in video action recognition, image super-resolution, and frame interpolation [2111.00772].

By contrast, "Comparison of Methods Generalizing Max- and Average-Pooling" reports that none of the more sophisticated pooling methods perform significantly better than standard max- or average-pooling in VGG16 on Google Open Images v5 [2103.01746]. Test accuracy values cluster near 39% for AP, NN, OP, LNP, and SEMP, while SMP variants are lower. This paper is an important counterexample to any claim that hybrid pooling is intrinsically superior.

A different form of HPD, defined as dynamic interleaving of full-resolution and low-mode passes with downsampled inputs and pooled kernels during training, reports efficiency rather than accuracy gains. On CIFAR-10, ResNet-18 shows 18% total time reduction with top-1 error moving from 6.7% to 6.9%, and ResNet-50 shows 23% total time reduction with error moving from 6.3% to 6.4% [1910.06548]. This indicates that HPD can also be used as a training-time efficiency device rather than only as an inference-time architectural module.

## 5. Theoretical properties, information retention, and computational trade-offs

The main theoretical argument across CNN formulations is information preservation under downsampling. In the DiffStride–Spectral formulation, fixed stride is said to miss features because it samples on a rigid grid, while max pooling imposes hard selection and a constraining lower bound on preserved information; spectral truncation is motivated by retaining low-frequency energy that carries global structure, and learnable stride adapts the sampling rate to the data [2401.09008]. The same paper references prior work by Rippel et al. for the claim that spectral pooling reduces the constraint on the lower bound of preserved information for a given output dimensionality, although explicit bounds are not proved there [2401.09008].

The MinMaxPooling variant bases its theory on “information complementarity”: maximum values emphasize bright structures and strong edges, minimum values emphasize dark structures, baseline, and depressions, and their fusion encodes local contrast and detail [2507.14790]. The paper explicitly notes that hard min/max yields sparse gradients, whereas soft min/max with LogSumExp gives dense gradients at the cost of approximation bias [2507.14790]. AdaPool advances a related but distinct argument: both of its exponential kernels produce strictly positive normalized weights, so all positions receive non-zero gradients, and the learned $\beta(R)$ adapts locally between a smooth-average-like and smooth-max-like regime [2111.00772].

The graph-coarsening formulation provides the clearest structural guarantees. The coarsened graph preserves the number of connected components, and for any coarsened edge $(u,v)$ with $u\neq v$ the original graph distance satisfies
$$
k+1 \le d_G(u,v) \le 2k+1.
$$
For centroid pairs in the same component, the paper further gives path-length bounds
$$
(k+1)\cdot d_H(i,j)\le d_G(i,j)\le (2k+1)\cdot d_H(i,j).
$$
These statements formalize distortion control under graph downsampling [2208.03523].

The computational trade-offs are equally consistent across papers. FFT and Hartley transforms introduce additional cost relative to max or average pooling in the DiffStride–Spectral method [2401.09008]. MinMaxPooling requires approximately two pooling passes plus optional $1\times 1$ convolution, and the paper reports modest or small FLOP and parameter increases depending on the backbone [2507.14790]. AdaPool reports minor additional parameter and FLOP overhead but clearly higher pooling latency than average pooling, with CUDA timings of 14 ms / 76 ms for avg pooling forward/backward versus 119 ms / 490 ms for AdaPool [2111.00772]. The graph coarsening method emphasizes $O(k|E|)$ work using parallel neighborhood reductions without explicitly materializing the dense $G^k$ graph [2208.03523]. The training-time HPD scheme predicts reduced per-iteration cost through simultaneous input downsampling and kernel pooling, with empirical low-mode/full-mode cost ratios around 0.37–0.39 for ResNet-18/50 at downsampling factor $r=2$ [1910.06548].

## 6. Architectural integration, limitations, and recurring design choices

Architecturally, HPD is usually inserted at canonical downsampling sites. In ResNet-18, the DiffStride–Spectral variant replaces strided convolutions in shortcut downsampling blocks, applies DiffStride to both main and residual branches for shape alignment, and places Spectral Pooling one level above GAP [2401.09008]. In UNet-like encoders, the MinMaxPooling variant replaces each downsampling operation with HPD using the standard $k=2$, $s=2$ pattern, while keeping skip connections unchanged [2507.14790]. In graph classification pipelines, hybrid pooling is applied after $K$ message-passing layers to produce graph-level embeddings, and Manifold-Mixup is performed at that graph-level representation rather than on adjacency structure [2210.03123]. In the graph-coarsening framework, HPD naturally occupies the coarsen-and-pool step between GNN blocks [2208.03523].

Several recurring design choices appear across the literature. Small or standard reduction factors are preferred: $k=2$, $s=2$ in segmentation encoders [2507.14790], $r=2$ in dynamic input and kernel downsampling [1910.06548], and moderate $k$ in graph coarsening for a balance between compression and distortion [2208.03523]. Hybrid fusion is also typically lightweight: convex combinations, concatenation plus $1\times 1$ projection, or per-region scalar masks rather than large auxiliary networks [2210.03123], [2111.00772], [2507.14790].

The limitations are equally recurrent. Statistical significance is not reported for the DiffStride–Spectral experiments [2401.09008]. The graph-learning study explicitly notes threats to external and construct validity, including dependence on UPFD and Project_CodeNet and evaluation robustness based on random edge dropping rather than other perturbation models [2210.03123]. The segmentation formulation warns that min pooling can be sensitive to outlier noise and that architectures already rich in multi-scale fusion, such as DeepLabv3+, may experience redundancy [2507.14790]. The comparison study on generalized pooling finds no significant advantage over standard max or average pooling in a large-scale image-classification setting, and observes that freely learned temperatures in Smooth-Maximum Pooling can destabilize training or drift toward overly sharp regimes [2103.01746]. AdaPool, while broadly successful, still adds measurable runtime overhead [2111.00772].

Taken together, these results rule out a simple interpretation of HPD as a universally dominant replacement for conventional downsampling. The literature instead supports a narrower conclusion: hybridization is beneficial when its chosen components match the information bottleneck of the task—frequency truncation for late CNN representations, min–max complementarity for medical image boundaries, adaptive exponential weighting for detail-preserving vision pipelines, or fused graph readouts for mixup-based graph classification. Where that match is weak, standard pooling may remain competitive or preferable.

Source: https://www.emergentmind.com/topics/hybrid-pooling-downsampling-hpd