Research map / Probabilistic and scientific modeling / Causal discovery
Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning
Yuhang Liu, Zhen Zhang, Dong Gong, Erdun Gao, Biwei Huang, Mingming Gong, Anton van den Hengel, Kun Zhang, Javen Qinfeng Shi
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 |
|---|---|---|---|
| apply FastICA to CLIP features before linear probe | Linear Probe (RN50) · ImageNet 2-shot | accuracy (%): 31.95 -> 34.06 | Improved |
| apply PCA then FastICA to CLIP features | Linear Probe (ViT16) · ImageNet 2-shot | accuracy (%): 44.97 -> 46.57 | Improved |
| apply PCA then FastICA to CLIP features | Linear Probe (ViT16) · ImageNet distribution shift (V2/Sketch/R/A avg) | accuracy (%): 32.51 -> 37.13 | Improved |
| add PCA before FastICA (RN101) (ablation) | Linear Probe + FastICA · ImageNet shift avg | accuracy (%): 22.93 -> 25.29 | Improved |
| add PCA before FastICA (ViT32) (ablation) | Linear Probe + FastICA · ImageNet 2-shot source | accuracy (%): 40.21 -> 39.34 | Hurt |
| FastICA on ImageNet-Sketch with RN50 | Linear Probe · ImageNet-Sketch | accuracy (%): 8.41 -> 8.37 | No clear effect |
| violate distribution assumptions on hypersphere (Laplace cond.) (ablation) | MMCL · Synthetic hypersphere | R^2 (%): 99.5 -> 99.4 | No clear effect |
| unbounded space with Normal model and Normal conditional (ablation) | MMCL · Synthetic, assumptions violated | R^2 (%): 99.5 -> 90.5 | Hurt |
| add FastICA to Tip-Adapter and Tip-Adapter-F | Tip-Adapter · 11 few-shot datasets | accuracy | Improved |
Closest papers by meaning
- RePercENT: Scaling Disentangled Representation Learning Beyond Two Modalities (NeurIPS 2026 · similarity 0.89)
- Causal Representation Learning from Multimodal Biomedical Observations (ICLR 2025 · similarity 0.88)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.88)
- An Information Criterion for Controlled Disentanglement of Multimodal Data (ICLR 2025 · similarity 0.88)
- Multi-View Causal Representation Learning with Partial Observability (ICLR 2024 · similarity 0.88)
- Disentangled Representation Learning in Non-Markovian Causal Systems (NeurIPS 2024 · similarity 0.88)
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.
