Jax-Lean Translator
A tool that translates Jax tensor computation graphs into Lean expressions, intended to use the Lean theorem prover for formal verification of machine learning code.
Timeline (last 90 days)
A work titled "Provably Correct Tensor Puzzles" built a Jax-to-Lean translator for the formal verification of Python machine learning code.
The work uses "tensor puzzles" as test cases to verify the correctness of the Jax-Lean translator.
The current work most likely covers only deterministic tensor operations over integers or exact rational numbers, while floating point and randomness remain open problems to be tackled.
If the Jax-Lean translator matures, developers can obtain formal guarantees while writing Jax code, which is especially applicable to safety-critical AI applications.
All Facts (4)
A work titled "Provably Correct Tensor Puzzles" built a Jax-to-Lean translator for the formal verification of Python machine learning code.
50%UnverifiedThe work uses "tensor puzzles" as test cases to verify the correctness of the Jax-Lean translator.
50%UnverifiedThe current work most likely covers only deterministic tensor operations over integers or exact rational numbers, while floating point and randomness remain open problems to be tackled.
50%UnverifiedIf the Jax-Lean translator matures, developers can obtain formal guarantees while writing Jax code, which is especially applicable to safety-critical AI applications.
50%