Research map / Probabilistic and scientific modeling / Causal discovery
Counterfactual Maximum Likelihood Estimation for Training Deep Networks
Xinyi Wang, Wenhu Chen, Michael Saxon, William Yang Wang
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 |
|---|---|---|---|
| train with Implicit or Explicit CMLE instead of MLE | MLP · simulated data OOD2 | MSE: 7.25 -> 6.94 | Improved |
| Explicit CMLE with p(T|X,Y) trained on interventional distribution (Explicit CMLE*) (ablation) | Explicit CMLE · simulated counterfactual test | MSE: 4.77 -> 4.61 | Improved |
| Explicit instead of Implicit CMLE (ablation) | CMLE · simulated counterfactual test | MSE: 5.11 -> 4.77 | Improved |
| CMLE vs MLE on in-distribution observational test | CMLE · simulated observational test | MSE: 3.63 -> 3.45 | Improved |
| fine-tune RoBERTa on counterfactual-augmented data with Implicit CMLE | RoBERTa · ANLI | accuracy: 31.3 -> 32.2 | Improved |
| augment with counterfactual data using plain MLE generator (ablation) | RoBERTa · ANLI | accuracy: 31.3 -> 30.8 | Hurt |
| CMLE augmentation on in-domain NLI | RoBERTa · SNLI | accuracy: 92.6 -> 92.4 | No clear effect |
| CMLE generators vs MLE for hypothesis label consistency | BART · SNLI generated hypotheses (NLI-RoBERTa) | label consistency: 65.5 -> 95.2 | Improved |
| Explicit instead of Implicit CMLE for generation consistency (ablation) | BART · MNLI generated hypotheses (NLI-RoBERTa) | label consistency: 75.1 -> 85.8 | Improved |
| train captioner on hallucination-augmented data with plain MLE (ablation) | Transformer captioner · MSCOCO Karpathy test | CIDEr: 106.7 -> 102.1 | Hurt |
| Implicit CMLE vs MLE on hallucination | Transformer captioner · MSCOCO Karpathy test | CHAIRs: 7.3 -> 6.8 | Improved |
| Implicit CMLE vs MLE on similarity metrics | Transformer captioner · MSCOCO Karpathy test | CIDEr: 106.7 -> 105.2 | No clear effect |
Closest papers by meaning
- Counterfactual Generative Modeling with Variational Causal Inference (ICLR 2025 · similarity 0.87)
- From Black-box to Causal-box: Towards Building More Interpretable Models (NeurIPS 2025 · similarity 0.86)
- Towards Characterizing Domain Counterfactuals for Invertible Latent Causal Models (ICLR 2024 · similarity 0.86)
- CF-OPT: Counterfactual Explanations for Structured Prediction (ICML 2024 · similarity 0.85)
- Exogenous Matching: Learning Good Proposals for Tractable Counterfactual Estimation (NeurIPS 2024 · similarity 0.85)
- Conditional Counterfactual Mean Embeddings: Doubly Robust Estimation and Learning Rates (NeurIPS 2026 · similarity 0.85)
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.
