GrootVL: Tree Topology is All You Need in State Space Model

TL;DR

GrootVL introduces a dynamic tree topology in state space models to improve long-range dependency modeling, achieving superior performance on visual and textual tasks with efficient computation.

Contribution

The paper proposes GrootVL, a novel state space model with dynamic tree topology generation and a linear complexity algorithm for enhanced long-range interactions.

Findings

Outperforms existing structured state space models on image tasks

Improves textual task performance with minimal fine-tuning cost

Achieves strong representation capabilities across modalities

Abstract

The state space models, employing recursively propagated features, demonstrate strong representation capabilities comparable to Transformer models and superior efficiency. However, constrained by the inherent geometric constraints of sequences, it still falls short in modeling long-range dependencies. To address this issue, we propose the GrootVL network, which first dynamically generates a tree topology based on spatial relationships and input features. Then, feature propagation is performed based on this graph, thereby breaking the original sequence constraints to achieve stronger representation capabilities. Additionally, we introduce a linear complexity dynamic programming algorithm to enhance long-range interactions without increasing computational cost. GrootVL is a versatile multimodal framework that can be applied to both visual and textual tasks. Extensive experiments demonstrate that our method significantly outperforms existing structured state space models on image classification, object detection and segmentation. Besides, by fine-tuning large language models, our approach achieves consistent improvements in multiple textual tasks at minor training cost.