Postdoc: Formal Verification for Verified ML Inference

EPFL; Lausanne; Vor Ort; CHF 110.000 - 140.000; Vollzeit

Benefits dieser Stelle

International working environment

Excellent working conditions

Frontier AI models access

Travel for collaboration

Zusammenfassung

The Keystone project at EPFL in Lausanne seeks exceptional candidates in formal methods and computer systems to advance machine-verified ML inference engines. You will contribute to GPU semantics, the coordination layer, and AI-assisted proof workflows, collaborating with Imperial College London on a high-profile ARIA effort.

Applicants should hold a PhD (or near completion) in CS with strong publication records and proficiency in Python, C++, OCaml, and related tools.

Qualifikationen

• PhD (or nearing completion) in computer science or closely related field.
• Background in formal verification, programming languages, systems, or ML.
• Research experience in interactive theorem proving (Rocq, Lean, HOL, Isabelle, or similar), GPU programming/semantics, compilers, concurrency, or ML systems.
• Strong software engineering across multiple languages (e.g., Python, C++, OCaml).
• Publication record relative to career stage in international venues.

Aufgaben

• Conduct research related to the Keystone project, including designing and building a machine-verified LLM inference engine.
• Explore formalizing GPU kernel semantics in a proof assistant and verify high-performance inference kernels.
• Develop and verify the inference coordination layer (batching, KV-cache, scheduling) with extraction to executable code.
• Develop agentic AI workflows for specification autoformalization, proof generation, and proof repair.
• Engage with ARIA programme through red/blue team exercises and sprint reviews.

Kenntnisse

Formal verification

Programming languages

Python

C++

OCaml

Lean/Rocq

AI/ML systems

Ausbildung

PhD in CS

Tools

Rocq

Lean

HOL

Isabelle

Jobbeschreibung

The Keystone project at EPFL in Lausanne seeks exceptional candidates in formal methods and computer systems to advance machine-verified ML inference engines. You will contribute to GPU semantics, the coordination layer, and AI-assisted proof workflows, collaborating with Imperial College London on a high-profile ARIA effort.

Applicants should hold a PhD (or near completion) in CS with strong publication records and proficiency in Python, C++, OCaml, and related tools.