Overview
Field: Machine Learning Authors: Levi Dubrovin, Nikita Pospelov, Kirill Sabitov Published: 2026-09-25 arXiv: 2609.27203
Summary
xlog is a CUDA-native logic programming engine integrating neural perception with deterministic Datalog, probabilistic inference, and epistemic world views through a typed frontend and provider-owned CUDA runtime. Its reasoning modes share device data planes, but their execution boundaries differ: ordinary Datalog and exact inference are host-orchestrated, while certified resident recursive and Monte Carlo sampled cores record zero tracked host-device transfers before a bounded terminal receipt. The probabilistic path supports end-to-end gradients through GPU knowledge compilation from provenance to CNF to Decision-DNNF, exact weighted model counting, and backward gradients. A final smoothed circuit is certified against its source formula before caching or evaluation.
Key Results
- Circuit caching: yields a 2.74x speedup on MNIST-addition training.
- Join subsystem: the worst-case-optimal join is 27.96x faster (geometric mean) than xlog's binary join baseline.
- Accuracy: MNIST-addition accuracy comparable to Scallop (0.9561 vs 0.9468).
- Triangle counting: across five hub-skewed cases, the Souffle-to-fused-xlog runtime ratio grows from 0.88x (Souffle faster at 150K edges) to 5.54x at 1.2M edges; fused mode peaks at 85-1,033 MB device memory versus 3,287-44,979 MB for materialization.
- Exact inference: correctness-equivalent to ProbLog2 but slower.
- Video benchmark: a proximity predicate trained purely via symbolic credit replaces hand-tuned geometric parameters with unchanged held-out accuracy; it fails ten-fold cross-validation on event-calculus rule search and does not transfer on leak-free splits.
- Marine corpus: weighted clauses beat hard selection with 0.065 F1, reproduced via a single chronological training run.