XLOG is a CUDA-native engine that integrates neural perception with Datalog and probabilistic inference. It achieves a 2.74x MNIST-addition training speedup and a 27.96x geometric-mean gain in worst-case-optimal joins over its baseline.
HOW THIS AFFECTS YOU
●
builderYou can use this engine to accelerate neurosymbolic reasoning workflows directly on the GPU.
●
researcherThis provides a framework for end-to-end gradients through GPU knowledge compilation.