Self-Attentive Pooling for Efficient Deep Learning

TL;DR

This paper introduces a novel non-local self-attentive pooling method that improves feature aggregation in CNNs, leading to higher accuracy and reduced memory usage, enabling deployment on resource-constrained devices.

Contribution

We propose a self-attentive pooling technique that captures non-local dependencies, outperforming existing pooling methods in accuracy and efficiency for deep learning models.

Findings

Surpasses state-of-the-art pooling techniques in accuracy on ImageNet.

Achieves up to 22x reduction in memory consumption during inference.

Improves MobileNet-V2 performance by 1.2% on ImageNet.

Abstract

Efficient custom pooling techniques that can aggressively trim the dimensions of a feature map and thereby reduce inference compute and memory footprint for resource-constrained computer vision applications have recently gained significant traction. However, prior pooling works extract only the local context of the activation maps, limiting their effectiveness. In contrast, we propose a novel non-local self-attentive pooling method that can be used as a drop-in replacement to the standard pooling layers, such as max/average pooling or strided convolution. The proposed self-attention module uses patch embedding, multi-head self-attention, and spatial-channel restoration, followed by sigmoid activation and exponential soft-max. This self-attention mechanism efficiently aggregates dependencies between non-local activation patches during down-sampling. Extensive experiments on standard object classification and detection tasks with various convolutional neural network (CNN) architectures demonstrate the superiority of our proposed mechanism over the state-of-the-art (SOTA) pooling techniques. In particular, we surpass the test accuracy of existing pooling techniques on different variants of MobileNet-V2 on ImageNet by an average of 1.2%. With the aggressive down-sampling of the activation maps in the initial layers (providing up to 22x reduction in memory consumption), our approach achieves 1.43% higher test accuracy compared to SOTA techniques with iso-memory footprints. This enables the deployment of our models in memory-constrained devices, such as micro-controllers (without losing significant accuracy), because the initial activation maps consume a significant amount of on-chip memory for high-resolution images required for complex vision tasks. Our proposed pooling method also leverages the idea of channel pruning to further reduce memory footprints.