Structural Conditional Independence Test via Neural Quantile Resolution
Organizations: Deakin Applied Artificial Intelligence Initiative, Geelong, Australia
Abstract
Conditional independence testing is difficult when dependence is weak, localized, or expressed through distributional changes rather than a global mean shift. Existing tests usually return a scalar statistic or p-value, which can detect rejection but gives limited resolution about where the conditional law changes. We propose QRCIT, a quantile-resolution conditional independence test that compares P(X | Z) and P(X | Y, Z) through an amortized quantile deformation. A lightweight residual quantile-flow model first learns the baseline conditional quantile law and then learns the additional deformation induced by Y, yielding both a calibrated global test statistic and an inspectable deformation field over context and quantile level. We also introduce the Structural Dependency Benchmark (SDB), which evaluates real-world-inspired dependence structures, such as context-local, tail-only, sign-reversing, moving-ridge, and mean-neutral patterns. Across synthetic PNL and SDB benchmarks, QRCIT improves detection of structured alternatives while maintaining empirical Type-I control. Increasing quantile resolution consistently improves merged-SDB power, and the local-rescue strategy sharpens detection when dependence is concentrated in structural regions. On Sachs, METABRIC breast cancer, and CCLE/PLX4720, QRCIT provides competitive real-data performance and interpretable deformation patterns. These results support quantile-resolved conditional-law deformation as a practical framework for testing and diagnosing structured conditional dependence.
Figures & tables
| Null | Alternative | |||
|---|---|---|---|---|
| 250 | 0.650 | 0.250 | 0.179 | 0.160 |
| 500 | 0.629 | 0.198 | 0.150 | 0.096 |
| 1000 | 0.655 | 0.082 | 0.067 | 0.052 |
| 250 | 0.60 | 0.61 | 0.60 |
|---|---|---|---|
| 500 | 0.70 | 0.73 | 0.75 |
| 1000 | 0.87 | 0.89 | 0.91 |
| 1500 | 0.98 | 0.98 | 0.98 |
| Metric | QRCIT | GCM | SGMCIT | CDCIT | KCI | PM-KCI | NNLSCIT |
|---|---|---|---|---|---|---|---|
| Type-I | 0.00 | 0.05 | 0.91 | 0.04 | 0.05 | 0.05 | N/A |
| Power | 1.00 | 1.00 | 1.00 | 0.24 | 1.00 | 1.00 | N/A |
| Orientation | Type-I | Power |
|---|---|---|
| Original | 0.10 | 0.99 |
| Swapped | 0.04 | 0.98 |
| Metric | QRCIT | NNLSCIT | GCM | SGMCIT | CDCIT |
|---|---|---|---|---|---|
| Precision | 0.733 | 0.746 | 0.676 | 0.776 | 0.700 |
| Recall | 0.880 | 0.940 | 0.920 | 0.760 | 0.840 |
| F1 | 0.800 | 0.832 | 0.780 | 0.768 | 0.764 |
| Q.10 | Q.05 | Q.01 | SGMCIT | CDCIT | NNLSCIT | GCM | |
|---|---|---|---|---|---|---|---|
| MSE | 0.965 | 0.972 | 0.974 | 0.977 | 0.981 | 0.985 | 0.990 |
| BRAF.V600E | BRAF.MC | HIP1 | CDC42BPA | THBS3 | |
|---|---|---|---|---|---|
| QRCIT | 0.014 | 0.008 | 0.075 | 0.028 | 0.901 |
| GCIT | 0.008 | 0.050 | 0.013 |
Conditional Independence Testing
Conditional independence testing asks whether two variables remain independent after accounting for other variables, supporting causal discovery and valid inference.
See all 23 papers