Summary of Coupled Mamba: Enhanced Multi-modal Fusion with Coupled State Space Model, by Wenbing Li et al.
Coupled Mamba: Enhanced Multi-modal Fusion with Coupled State Space Model
by Wenbing Li, Hang Zhou, Junqing Yu, Zikai Song, Wei Yang
First submitted to arxiv on: 28 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to multi-modal fusion is proposed by combining State Space Models (SSMs) with a Coupled SSM architecture, which leverages the strengths of both in capturing complex intra- and inter-modality correlations. This paper builds upon recent advancements in SSMs, particularly the Mamba model, and introduces an inter-modal hidden states transition scheme to enable modality fusion while maintaining independence of intra-modality state processes. The Coupled SSM model is designed for parallelism, incorporating an expedite coupled state transition scheme and a global convolution kernel. Experimental results on CMU-MOSEI, CH-SIMS, and CH-SIMSV2 datasets demonstrate the effectiveness of this approach, achieving improved F1-scores (0.4%, 0.9%, and 2.3%) and faster inference (49% reduction) while utilizing less GPU memory (83.7% savings). This work paves the way for enhanced multi-modal fusion in various applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper explores a new way to combine different types of information, like text, images, or audio, into a single representation that’s better than any one of them alone. The authors use special models called State Space Models (SSMs) and find a way to make them work together even when the information is complex and correlated across modalities. They test their approach on three datasets and show it performs better than existing methods while being faster and using less computer memory. |
Keywords
* Artificial intelligence * Inference * Multi modal
