Saved in:
Bibliographic Details
Main Authors: Sui, Yueyuan, Mohapatra, Payal, Eldenk, Doğaç, Yang, Haodong, Zhang, Yiting, Zhang, Haoyan, Zhu, Qi, Xia, Stephen
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.08971
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908951235264512
author Sui, Yueyuan
Mohapatra, Payal
Eldenk, Doğaç
Yang, Haodong
Zhang, Yiting
Zhang, Haoyan
Zhu, Qi
Xia, Stephen
author_facet Sui, Yueyuan
Mohapatra, Payal
Eldenk, Doğaç
Yang, Haodong
Zhang, Yiting
Zhang, Haoyan
Zhu, Qi
Xia, Stephen
contents Edge devices increasingly run multimodal sensing pipelines that must remain accurate despite fluctuating power budgets and unpredictable sensor dropout. Existing pruning methods fail under these conditions: they generally require fine-tuning after compression, consuming over $10\times$ the deployment energy, and they assign static importance scores that are blind to which sensors are present. We present the SentryFuse framework, which addresses both challenges jointly through two key components. First, SentryGate learns modality-conditioned importance scores during training via first-order saliency supervision and then prunes attention heads and feed-forward channels at deployment without fine-tuning. Second, SentryAttend replaces dense self-attention, a key bottleneck in contemporary multimodal architectures, with sparse grouped-query attention, yielding a net 15% reduction in GFLOPs across three different multimodal architectures. Across three applications and multimodal backbones, SentryGate achieves a 12.7% average accuracy improvement over the strongest pruning baseline, and upto to 18% under modality dropout conditions. Together, SentryFuse reduces memory by 28.2% and lowers latency by up to $1.63\times$ without further fine-tuning, establishing modality-aware zero-shot compression as a practical path to multimodal intelligence on heterogeneous edge hardware.
format Preprint
id arxiv_https___arxiv_org_abs_2604_08971
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Modality-Aware Zero-Shot Pruning and Sparse Attention for Efficient Multimodal Edge Inference
Sui, Yueyuan
Mohapatra, Payal
Eldenk, Doğaç
Yang, Haodong
Zhang, Yiting
Zhang, Haoyan
Zhu, Qi
Xia, Stephen
Machine Learning
Edge devices increasingly run multimodal sensing pipelines that must remain accurate despite fluctuating power budgets and unpredictable sensor dropout. Existing pruning methods fail under these conditions: they generally require fine-tuning after compression, consuming over $10\times$ the deployment energy, and they assign static importance scores that are blind to which sensors are present. We present the SentryFuse framework, which addresses both challenges jointly through two key components. First, SentryGate learns modality-conditioned importance scores during training via first-order saliency supervision and then prunes attention heads and feed-forward channels at deployment without fine-tuning. Second, SentryAttend replaces dense self-attention, a key bottleneck in contemporary multimodal architectures, with sparse grouped-query attention, yielding a net 15% reduction in GFLOPs across three different multimodal architectures. Across three applications and multimodal backbones, SentryGate achieves a 12.7% average accuracy improvement over the strongest pruning baseline, and upto to 18% under modality dropout conditions. Together, SentryFuse reduces memory by 28.2% and lowers latency by up to $1.63\times$ without further fine-tuning, establishing modality-aware zero-shot compression as a practical path to multimodal intelligence on heterogeneous edge hardware.
title Modality-Aware Zero-Shot Pruning and Sparse Attention for Efficient Multimodal Edge Inference
topic Machine Learning
url https://arxiv.org/abs/2604.08971