Research map / Architectures and efficiency / CNN design
Deep Multimodal Multilinear Fusion with High-order Polynomial Pooling
Ming Hou, Jiajia Tang, Jianhai Zhang, Wanzeng Kong, Qibin Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add hierarchical second PTP layer to single-layer HPFN (ablation) | HPFN · CMU-MOSI | Acc-7: 35.9 -> 36.3 | Improved |
| stack to four layers instead of two (ablation) | HPFN-L2 · CMU-MOSI | Acc-7: 36.3 -> 34.6 | Hurt |
| add dense skip connections of unimodal features to two-layer HPFN (ablation) | HPFN-L2 · CMU-MOSI | Acc-7: 36.3 -> 36.5 | Improved |
| add a third layer to two-layer HPFN (ablation) | HPFN-L2 · CMU-MOSI | Acc-7: 36.3 -> 36.0 | Hurt |
| use temporal overlapped windows instead of time-averaged non-temporal features (ablation) | HPFN-L2 · CMU-MOSI | MAE: 0.958 -> 0.944 | Improved |
| share a single PTP across temporal windows (ablation) | temporal HPFN-L2 · CMU-MOSI | Acc-7: 36.7 -> 35.7 | Hurt |
| fuse all three modalities instead of text only (ablation) | HPFN P=[4] · CMU-MOSI | Acc-7: 35.1 -> 36.0 | Improved |
| vary polynomial order from 1 to 10 (ablation) | HPFN · CMU-MOSI | Acc-2/F1 | Mixed |
| use HPFN-L2 instead of LMF | IEMOCAP · IEMOCAP | F1-Angry: 89.0 -> 88.8 | Hurt |
| use HPFN-L2 instead of LMF | CMU-MOSI · CMU-MOSI | MAE: 0.912 -> 0.945 | Hurt |
Closest papers by meaning
- Convolutions and More as Einsum: A Tensor Network Perspective with Advances for Second-Order Methods (NeurIPS 2024 · similarity 0.85)
- BLOCK: Bilinear Superdiagonal Fusion for Visual Question Answering and Visual Relationship Detection (AAAI 2019 · similarity 0.85)
- Multilinear Operator Networks (ICLR 2024 · similarity 0.83)
- GeminiFusion: Efficient Pixel-wise Multimodal Fusion for Vision Transformer (ICML 2024 · similarity 0.83)
- Coupled Mamba: Enhanced Multimodal Fusion with Coupled State Space Model (NeurIPS 2024 · similarity 0.83)
- Deep Tree Tensor Networks (NeurIPS 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
