GFlowQEC generative flow networks sample quantum LDPC codes more effectively
Original title: GFlowQEC: Generative Flow Networks that Learn to Sample Quantum LDPC Codes
GFlowQEC finds more than twice as many top-scoring codes as parallel tempering when ranking ten draws per paid evaluation.
GFlowQEC is the first generative flow network applied to quantum code design. It builds codes one group element at a time and learns to propose codes in proportion to a power of their score. Trained once on scores from earlier searches and then frozen, it proposes lifted-product codes with 180 physical qubits. When ranking ten draws per paid call it finds more than twice as many new top-scoring codes inequivalent to its training set than tuned parallel tempering and more than three and a half times as many as a supervised screen.
- Trained once on scores from earlier searches
- Proposes codes with 180 physical qubits
- More than twice as many codes as parallel tempering
- More than three and a half times as many as supervised screen
- Soft Q-learning performs similarly
From the original article
1 min read at arxiv.orgChoosing a quantum error-correcting code for a real device means scoring candidates with circuit-level simulations so slow that only a few hundred can be run. So every evaluation counts, and a search is judged by how many good codes it finds per paid evaluation. Quantum LDPC codes built from a few group elements protect many logical qubits, but good ones are rare in a huge space.
By Nuno Batista · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.