Internship: Parameter optimization for message-passing qLDPC decoders

Background

Message-passing decoders for qLDPC codes, such as belief propagation and its variants (for example relay-BP), have many tunable parameters: memory and damping strengths, iteration counts, number and order of relay stages, and stopping rules. Logical error rate and decoding time depend strongly on these choices. Each evaluation needs a large number of Monte Carlo samples, so a brute-force search costs too much.

Project

The intern will review and compare state-of-the-art methods for tuning these parameters on bivariate bicycle codes under circuit-level noise:

• Random sampling as a baseline.
• Successive halving and Hyperband, which spend samples on promising configurations and drop the rest early.
• Submodular optimization, for choosing a small set of complementary decoder configurations (for example the stages of a relay or ensemble decoder), with greedy algorithms that carry approximation guarantees.

Deliverables are a benchmark of the methods by logical error rate, decoding time, and tuning cost, and a set of recommended parameter configurations.

Requirements

• Master's or PhD student in computer science, applied mathematics, physics, electrical engineering, or a related field.
• Good Python skills.
• Background in optimization, statistics, or machine learning.

Nice to have

• Knowledge of belief propagation or classical LDPC decoding.
• Experience with hyperparameter optimization tools (Optuna, Ray Tune).
• Basic knowledge of quantum error correction.