DeepSWIP: Quotient-WMC Counterfactuals for Neural Probabilistic Logic Programs

2026-06-18

AI Summary

Existing neuro-symbolic systems, such as DeepProbLog, lack causal semantics and thus struggle to support counterfactual reasoning. This work proposes DeepSWIP, which introduces single-world counterfactual semantics into DeepProbLog for the first time. By employing neural instantiation to translate neural predicates into standard ProbLog choices, DeepSWIP integrates the Single-World Intervention Protocol (SWIP) with Weighted Model Counting (WMC) within a unified framework to compute counterfactuals exactly. The approach reveals a quotient-based WMC mechanism underlying neural probabilistic activations and clarifies key phenomena including intervention cleanup, calibration sensitivity, and instability under rare evidence. Empirical evaluation demonstrates that DeepSWIP correctly validates 12,000 queries on MPI3D with a 2.14× speedup in inference, while experiments on SUMO HOV further confirm its ability to substantially reduce bias in plug-in estimators.

Abstract

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.

Problem

• counterfactual reasoning
• neurosymbolic systems
• causal semantics
• interventions
• probabilistic logic

Innovation

• counterfactual reasoning
• neurosymbolic AI
• weighted model counting
• single-world intervention
• quotient-WMC