Answers are grounded in this paper and cite their sources.
Demonstrates correctness in symbolic probabilistic languages, suggesting improved debugging and optimization.
Symbolic execution has emerged as a powerful technique for scaling exact probabilistic inference to languages with more expressive features. But, this expressivity comes at a price: probabilistic programming languages based on symbolic execution are difficult to debug, optimize, and prove correct due to the many intricacies inherent to high-performance symbolic execution strategies. We aim to make it easier to work with probabilistic symbolic executors by developing symbolic sets, a new semantic domain that cleanly captures the notion of computation underlying symbolic execution. Just as a symbolic executor replaces ordinary execution with a lifted semantics, symbolic set theory replaces ordinary set theory with a lifted mathematics: the category of symbolic sets is a Grothendieck topos, which allows type theory to be used as a metalanguage for working with symbolic sets and functions. We prove a metatheorem that shows how a large class of definitional interpreters written in the internal language of symbolic sets are automatically correct for their ordinary set-theoretic interpretations. Using this metatheorem, we give the first full correctness argument for a symbolic probabilistic language with higher-order functions, type-directed state merging, pattern matching, and structural recursion.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: