Users in the replies praise the Anandkumar Group's BRIDGE decomposition into code/spec/proof domains for its ICML papers because direct natural language to Lean translation was always going to hit a wall.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@AnimaAnandkumar @icmlconf BRIDGE decomposing into code/spec/proof domains is the right call. Direct NL-to-Lean was always going to hit a wall with Lean's total/immutable recursion style. Functional intermediate reps bridge that gap naturally.
Excited to share four Lean-related papers from our group at @icmlconf workshops in Math and Physics! Together, these works explore how theorem proving can support verified ML systems, functional program synthesis, interoperability across proof assistants, and scientific reasoning. A common theme is using Lean not just as a proof assistant for mathematics, but as infrastructure for building, specifying, checking, and evaluating AI systems. @Robertljg, Jennifer Cruden, Will Adkisson, UIUC team (Xiangru Zhong, @huan_zhang12), @AmazonScience Team (Carson Eisenach, @udayaghai, Dominique Perrault-Joncas, Dean Foster), Jiayi Wu, Isha Goswami, and Anushka Paulchoudhury. @Caltech #ICML2026 #AI4Math #AI4Physics #Lean #FormalVerification #AI4Science #MachineLearning
QuantumLean-Bench appears at the @icmlconf 2026 AI4Physics Workshop and evaluates quantum-science reasoning with both informal explanations and Lean-oriented formal outputs. Quantum reasoning is a strong testbed for AI4Science because it combines physics, linear algebra, probability, circuits, states, operators, and domain-specific notation. The benchmark asks whether models can move from natural-language reasoning toward structured, machine-checkable scientific representations. Paper: https://openreview.net/forum?id=K9bHv3Y3Uf Repo: https://github.com/lean-dojo/QuantumLean-Bench
Users in the replies praise the Anandkumar Group's BRIDGE decomposition into code/spec/proof domains for its ICML papers because direct natural language to Lean translation was always going to hit a wall.
Based on 1 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.