Spectral-Progressive Thought Flow for Lightweight Multimodal Reasoning

2026-06-01

AI Summary

This work addresses the high computational and memory overhead in multimodal spatial reasoning caused by long chains of intermediate visual and textual reasoning steps. To mitigate this, the authors propose a lightweight inference framework that encodes intermediate visual states into fixed-size discrete cosine spectral representations. A streaming update mechanism aligns these representations with textual intent, while a spectral progressive strategy preserves global layout and injects high-frequency details on demand. Classifier-free guidance drives visual state updates, effectively preventing context inflation. Experimental results demonstrate that the proposed framework achieves comparable or superior reasoning performance while reducing computational cost and KV cache overhead by up to 2.1×, yielding stable latency and memory consumption independent of reasoning depth.

Abstract

Multimodal spatial reasoning often relies on long chains of intermediate textual and visual thoughts, where accumulating visual tokens and dense cross-modal attention incur substantial computation and memory overhead. To address this challenge, we propose Spectral-Progressive Thought Flow (SpecFlow), a novel lightweight multimodal spatial reasoning framework that represents intermediate visual thoughts in a fixed-size discrete cosine space. By exploiting strong energy compaction, SpecFlow preserves global layout and relational structure while introducing high-frequency details only when increased spatial precision is required. To align visual state evolution with linguistic intent, classifier-free guidance enables autoregressive textual thoughts to steer flow-based updates of the visual workspace/state without expanding the context. As a result, SpecFlow maintains a bounded visual workspace whose updates depend only on the current visual state and accumulated textual trace, enabling long-horizon inference with stable latency and memory usage independent of reasoning depth. Empirical results show that SpecFlow achieves competitive or superior reasoning performance while reducing computation and KV cache costs by up to 2.1 times.

Problem

• multimodal reasoning
• visual tokens
• cross-modal attention
• computation overhead
• memory overhead

Innovation

• Spectral Representation
• Lightweight Multimodal Reasoning
• Discrete Cosine Transform
• Classifier-Free Guidance
• Bounded Visual Workspace