Arcee: Differentiable Recurrent State Chain for Generative Vision Modeling with Mamba SSMs

TL;DR

Arcee introduces a differentiable recurrent state chain for vision modeling with Mamba SSMs, enabling cross-block memory reuse and significantly improving unconditional image generation quality.

Contribution

It proposes a novel cross-block recurrent state chain that reuses terminal state representations, compatible with existing Mamba variants, and enhances generative performance.

Findings

Reduces FID from 82.81 to 15.33 on CelebA-HQ

Compatible with all prior vision-mamba variants

Provides constant, negligible computational overhead

Abstract

State-space models (SSMs), Mamba in particular, are increasingly adopted for long-context sequence modeling, providing linear-time aggregation via an input-dependent, causal selective-scan operation. Along this line, recent "Mamba-for-vision" variants largely explore multiple scan orders to relax strict causality for non-sequential signals (e.g., images). Rather than preserving cross-block memory, the conventional formulation of the selective-scan operation in Mamba reinitializes each block's state-space dynamics from zero, discarding the terminal state-space representation (SSR) from the previous block. Arcee, a cross-block recurrent state chain, reuses each block's terminal state-space representation as the initial condition for the next block. Handoff across blocks is constructed as a differentiable boundary map whose Jacobian enables end-to-end gradient flow across terminal boundaries. Key to practicality, Arcee is compatible with all prior "vision-mamba" variants, parameter-free, and incurs constant, negligible cost. As a modeling perspective, we view terminal SSR as a mild directional prior induced by a causal pass over the input, rather than an estimator of the non-sequential signal itself. To quantify the impact, for unconditional generation on CelebA-HQ (256$\times$256) with Flow Matching, Arcee reduces FID$\downarrow$ from $82.81$ to $15.33$ ($5.4\times$ lower) on a single scan-order Zigzag Mamba baseline. Efficient CUDA kernels and training code will be released to support rigorous and reproducible research.