Modality-Aware Zero-Shot Pruning and Sparse Attention for Efficient Multimodal Edge Inference

TL;DR

The paper introduces SentryFuse, a framework for modality-aware zero-shot pruning and sparse attention that improves efficiency and robustness of multimodal models on edge devices without fine-tuning.

Contribution

It proposes a novel approach combining modality-conditioned importance learning and sparse attention to enable zero-shot pruning and efficiency gains.

Findings

Achieves 12.7% average accuracy improvement over baseline pruning methods.

Reduces memory usage by 28.2% and latency by up to 1.63 times.

Yields 15% reduction in GFLOPs across multiple architectures.

Abstract

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.