DeepSWIP: Quotient-WMC Counterfactuals for Neural Probabilistic Logic Programs
This work provides a practical method for counterfactual reasoning in neurosymbolic AI, addressing a key limitation of associational inference in systems like DeepProbLog.
DeepSWIP introduces a single-world counterfactual semantics for DeepProbLog programs, enabling exact counterfactual reasoning via weighted model counting. Experiments show a 2.14× speedup over the Twin network approach and demonstrate that neural calibration degradation biases plug-in estimates, which can be corrected with a properly scoped AIPW estimator.
Neurosymbolic systems such as DeepProbLog combine neural perception with probabilistic logic, but standard inference is associational. Counterfactual reasoning additionally requires a causal semantics for interventions and evidence. We introduce DeepSWIP, a single-world counterfactual semantics for DeepProbLog programs. Using neural materialization, we reduce fixed-context neural predicates to ordinary ProbLog choices, apply Single World Intervention Programs (SWIPs), and compute counterfactuals by weighted model counting (WMC) over a single transformed program. Under finite grounding and unique-supported-model assumptions, DeepSWIP is exact relative to the learned materialized FCM. The standard quotient-WMC form of ProbLog conditionals identifies active neural probabilities and explains intervention cleaning, calibration sensitivity, and rare-evidence instability. Experiments on MPI3D confirm the transformation against a DeepTwin construction against 12,000 queries, as predicted and a 2.14$\times$ inference speedup from avoiding the Twin's endogenous duplication. A SUMO HOV experiment shows that neural calibration degradation biases plug-in estimates, while a correctly scoped randomized-policy AIPW estimator removes most first-order bias for population mean and ATE estimands. Code is at https://github.com/saibib/deep_SWIP.