FOAA: Flattened Outer Arithmetic Attention for Multimodal Tumor Classification

Our Contributions

• We propose a Flattened Outer Arithmetic Fusion (FOAA) mechanism, which introduces novel ways to compute attention scores with four operations: outer addition (OA), outer product (OP), outer subtraction (OS), and outer division (OD). These operations are simple and reusable.
• We employ a self-attention mechanism in FOAA to extract discriminative features from learned feature maps that further improve the performance of classifiers for single modality tasks.
• We apply FOAA in a cross-attention mechanism, achieving state-of-the-art multi-modality fusion on multiple datasets.

Paper Available Here