Research map / Probabilistic and scientific modeling / Causal discovery
Learning Robust Intervention Representations with Delta Embeddings
Panagiotis Alimisis, Christos Diou
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 delta embeddings with CDE losses instead of vanilla classifier | ViT-DINO · Causal Triplet single-object ProcTHOR (OOD systematic) | accuracy: 0.47 -> 0.75 | Improved |
| add supervised contrastive loss (ablation) | CDE ViT-DINO · Causal Triplet single-object ProcTHOR (OOD systematic) | accuracy: 0.68 -> 0.75 | Improved |
| add sparsity loss (ablation) | CDE ViT-DINO · Causal Triplet single-object ProcTHOR (OOD systematic) | accuracy: 0.73 -> 0.75 | Improved |
| add contrastive and sparsity losses to delta embedding trained with CE (ablation) | CDE ViT-DINO · Causal Triplet single-object ProcTHOR (OOD systematic) | accuracy: 0.67 -> 0.75 | Improved |
| use delta embedding with CE only vs vanilla baseline (ablation) | ViT-DINO · Causal Triplet single-object ProcTHOR (OOD systematic) | accuracy: 0.47 -> 0.67 | Improved |
| use patch-wise Top-K model with larger k (ablation) | patch-wise CDE · multi-object ProcTHOR (smaller subset) | OOD accuracy: 0.42 -> 0.48 | Improved |
| increase k on real-world scenes (ablation) | patch-wise CDE · Epic-Kitchens (smaller subset) | OOD accuracy: 0.12 -> 0.15 | Improved |
| larger alpha_contrast than alpha_sparsity (ablation) | CDE · Causal Triplet single-object ProcTHOR (OOD systematic, smaller subset) | accuracy: 0.31 -> 0.33 | Improved |
| use CDE with ResNet-18 backbone (ablation) | CDE · single-object systematic (smaller subset) | accuracy: 0.54 -> 0.45 | Hurt |
| use ViT-MAE vs ViT-DINO backbone on multi-object | patch-wise CDE · Causal Triplet multi-object ProcTHOR (systematic shift) | OOD accuracy: 0.45 -> 0.30 | Hurt |
| CDE vs oracle-mask baseline | patch-wise CDE ViT-CLIP · Causal Triplet multi-object ProcTHOR (systematic shift) | OOD accuracy: 0.42 -> 0.48 | Improved |
| CDE vs best baseline on real-world Epic-Kitchens | patch-wise CDE ViT-CLIP · Causal Triplet Epic-Kitchens (systematic shift) | OOD accuracy: 0.27 -> 0.34 | Improved |
Closest papers by meaning
- A Sparsity Principle for Partially Observable Causal Representation Learning (ICML 2024 · similarity 0.89)
- Causal Representation Learning via Counterfactual Intervention (AAAI 2024 · similarity 0.88)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.87)
- Visual Representation Learning through Causal Intervention for Controllable Image Editing (CVPR 2025 · similarity 0.87)
- Multi-View Causal Representation Learning with Partial Observability (ICLR 2024 · similarity 0.87)
- Causal Representation Learning Made Identifiable by Grouping of Observational Variables (ICML 2024 · similarity 0.87)
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.
