# Gated Recursive Fusion: A Stateful Approach to Scalable Multimodal Transformers

arXiv: https://arxiv.org/abs/2507.02985 · HTML: https://mlsanity.com/paper/2507.02985/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 11 of its 16 references that are in the map.
- First posted on arXiv 1 Jul 2025.

## Builds on (most cited first, 11 of 11)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks](https://mlsanity.com/paper/1908.02265/) (2019)
- [Multimodal Machine Learning: A Survey and Taxonomy](https://mlsanity.com/paper/1705.09406/) (2017)
- [FLAVA: A Foundational Language And Vision Alignment Model](https://mlsanity.com/paper/2112.04482/) (2021)
- [Multimodal Transformer for Unaligned Multimodal Language Sequences](https://mlsanity.com/paper/1906.00295/) (2019)
- [Perceiver IO: A General Architecture for Structured Inputs & Outputs](https://mlsanity.com/paper/2107.14795/) (2021)
- [Tensor Fusion Network for Multimodal Sentiment Analysis](https://mlsanity.com/paper/1707.07250/) (2017)
- [MISA: Modality-Invariant and -Specific Representations for Multimodal Sentiment Analysis](https://mlsanity.com/paper/2005.03545/) (2020)
- [Memory Fusion Network for Multi-view Sequential Learning](https://mlsanity.com/paper/1802.00927/) (2018)
- [Improving Multimodal Fusion with Hierarchical Mutual Information Maximization for Multimodal Sentiment Analysis](https://mlsanity.com/paper/2109.00412/) (2021)

## Built on by (most cited first, 2 of 2)

- [Towards Explainable Adjudicative Variance: Quantifying Judicial Discretion via Gated Multi-Task Learning](https://mlsanity.com/paper/2606.27069/) (2026)
- [Grounding Foundational Vision Models with 3D Human Poses for Robust Action Recognition](https://mlsanity.com/paper/2511.05622/) (2025)

## Abstract (arXiv)

Multimodal learning faces a fundamental tension between deep, fine-grained fusion and computational scalability. While cross-attention models achieve strong performance through exhaustive pairwise fusion, their quadratic complexity is prohibitive for settings with many modalities. We address this challenge with Gated Recurrent Fusion (GRF), a novel architecture that captures the power of cross-modal attention within a linearly scalable, recurrent pipeline. Our method processes modalities sequentially, updating an evolving multimodal context vector at each step. The core of our approach is a fusion block built on Transformer Decoder layers that performs symmetric cross-attention, mutually enriching the shared context and the incoming modality. This enriched information is then integrated via a Gated Fusion Unit (GFU) a GRU-inspired mechanism that dynamically arbitrates information flow, enabling the model to selectively retain or discard features. This stateful, recurrent design scales linearly with the number of modalities, O(n), making it ideal for high-modality environments. Experiments on the CMU-MOSI benchmark demonstrate that GRF achieves competitive performance compared to more complex baselines. Visualizations of the embedding space further illustrate that GRF creates structured, class-separable representations through its progressive fusion mechanism. Our work presents a robust and efficient paradigm for powerful, scalable multimodal representation learning.