Research map / Probabilistic and scientific modeling / Causal discovery
Decoding Multimodal Causality: End-to-End Multimodal Mediation Pathways Inference
Yulong Li, Xiwei Liu, Niranjana Menon, Yuxuan Zhang, Jianxu Chen, Rong Xia, Haolin Yang, Peixin Guo, Yutong Xie, Imran Razzak
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 causal constraint loss Lcausal (ablation) | MM-SIGMA · Synthetic multimodal (end-to-end mediation) | pathway recovery: 0.61 -> 0.74 | Improved |
| add cross-modal alignment loss Lalign (ablation) | MM-SIGMA · Synthetic multimodal (end-to-end mediation) | pathway recovery: 0.65 -> 0.74 | Improved |
| add Bayesian model averaging (ablation) | MM-SIGMA · Synthetic multimodal (end-to-end mediation) | 95% CI coverage: 0.87 -> 0.94 | Improved |
| add Bayesian model averaging (ablation) | MM-SIGMA · Synthetic multimodal (end-to-end mediation) | pathway recovery: 0.71 -> 0.74 | Improved |
| lower path stability threshold from 0.5 to 0.3 (ablation) | MM-SIGMA · Synthetic multimodal (end-to-end mediation) | pathway recovery and bias: recovery 0.74 -> 0.79; NDE bias 0.012 -> 0.024; coverage 0.94 -> 0.91 | Mixed |
| raise path stability threshold from 0.5 to 0.7 (ablation) | MM-SIGMA · Synthetic multimodal (end-to-end mediation) | pathway recovery: 0.74 -> 0.68 | Hurt |
| multimodal adaptation instead of concatenation for NOTEARS | NOTEARS · Synthetic multimodal (structure discovery) | adjacency F1: 0.321 -> 0.389 | Improved |
| end-to-end MM-SIGMA instead of two-stage MM-DECI+EIF | MM-DECI+EIF · Synthetic multimodal (end-to-end mediation) | pathway recovery: 0.64 -> 0.74 | Improved |
| increase sample size from medium to large | MM-SIGMA · Synthetic multimodal (pathway recovery rate) | pathway recovery rate: 0.487 -> 0.512 | Improved |
| add strong observation noise | MM-SIGMA · Synthetic multimodal (pathway recovery rate) | pathway recovery rate: 0.487 -> 0.412 (15.4% degradation) | Hurt |
| add a second cross-modal transition to the pathway (S→I→S→I vs S→I→S) | MM-SIGMA · Synthetic multimodal (pathway recovery rate) | pathway recovery rate: 0.72 -> 0.63 | Hurt |
Closest papers by meaning
- Causal Representation Learning from Multimodal Biomedical Observations (ICLR 2025 · similarity 0.88)
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · similarity 0.87)
- Revealing Multimodal Causality with Large Language Models (NeurIPS 2025 · similarity 0.87)
- Controllable Generative Sandbox for Causal Inference (NeurIPS 2026 · similarity 0.86)
- When Additive Noise Meets Unobserved Mediators: Bivariate Denoising Diffusion for Causal Discovery (NeurIPS 2025 · similarity 0.86)
- Differentiable Causal Discovery for Latent Hierarchical Causal Models (ICLR 2025 · 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.
