Research map / Vision-language and multimodal / Cross-modal learning
Rethinking Gating Mechanism in Sparse MoE: Handling Arbitrary Modality Inputs with Confidence-Guided Gate
Liangwei Nathan Zheng, Wei Emma Zhang, Mingyu Guo, Olaf Maennel, Weitong Chen
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 |
|---|---|---|---|
| use ConfNet confidence gating instead of softmax (ablation) | ConfSMoE · MIMIC-III 48-IHM | F1: 48.59 -> 53.44 | Improved |
| use ConfNet confidence gating instead of softmax with load-balance loss (ablation) | ConfSMoE · MIMIC-III 48-IHM | F1: 51.67 -> 53.44 | Improved |
| use ConfNet instead of mean uniform expert weights (ablation) | ConfSMoE · MIMIC-III 48-IHM | F1: 48.34 -> 53.44 | Improved |
| use ConfNet instead of Gaussian gating (ablation) | ConfSMoE · MIMIC-III 48-IHM | F1: 46.50 -> 53.44 | Improved |
| use ConfNet instead of Laplacian gating (ablation) | ConfSMoE · MIMIC-III 48-IHM | F1: 50.10 -> 53.44 | Improved |
| use ConfNet instead of LFB gating (ablation) | ConfSMoE · MIMIC-III 48-IHM | F1: 48.36 -> 53.44 | Improved |
| use ConfNet instead of softmax (ablation) | ConfSMoE · CMU-MOSI 50% missing | AUC: 68.77 -> 70.41 | Improved |
| use ConfNet instead of softmax with load-balance loss (ablation) | ConfSMoE · CMU-MOSI 50% missing | F1: 42.43 -> 44.34 | Improved |
| add load-balance loss to softmax gating (ablation) | softmax router · MIMIC-III 48-IHM | F1: 48.59 -> 51.67 | Improved |
| add load-balance loss to softmax gating (ablation) | softmax router · CMU-MOSI 50% missing | AUC: 68.77 -> 67.40 | Hurt |
| Laplacian and Gaussian gates without load-balance loss vs softmax w/ load-balance (ablation) | gating function · MIMIC-III | F1: 46.50 -> 51.67 | No clear effect |
| add the two-stage imputation module (with ConfNet) (ablation) | ConfSMoE · MIMIC / CMU datasets | F1 and AUC | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FlexMoE — Flex-MoE: Modeling Arbitrary Modality Combination via the Flexible Mixture-of-Experts (NeurIPS 2024) · better on 13 of 13 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FuseMoE: Mixture-of-Experts Transformers for Fleximodal Fusion (NeurIPS 2024 · similarity 0.88)
- Flex-MoE: Modeling Arbitrary Modality Combination via the Flexible Mixture-of-Experts (NeurIPS 2024 · similarity 0.87)
- A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts (ICML 2024 · similarity 0.87)
- On Linear Mode Connectivity of Mixture-of-Experts Architectures (NeurIPS 2025 · similarity 0.87)
- MoDE: A Mixture-of-Experts Model with Mutual Distillation among the Experts (AAAI 2024 · similarity 0.86)
- Sigmoid Gating is More Sample Efficient than Softmax Gating in Mixture of Experts (NeurIPS 2024 · similarity 0.86)
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.
