Learning angles for quantum approximate optimization
Learned models predict high-quality QAOA angles in one pass, matching slower per-instance solvers.
The paper learns a mapping from combinatorial problem instances to high-quality variational angles for the Quantum Approximate Optimization Algorithm. A single forward pass replaces expensive per-instance classical optimization and is several orders of magnitude faster at nearly the same solution quality. The open-source models generalize to unseen Barabási-Albert and 2D grid graphs, and quantum-hardware samples using learned angles are indistinguishable from solver angles. Topology-sensitive models further improve angle quality, amortizing training when many similar instances are solved.
- Learned models output QAOA angles in one forward pass instead of per-instance optimization.
- Quality is nearly identical while inference is several orders of magnitude faster.
- Models generalize to unseen Barabási-Albert and 2D grid topologies.
- Hardware samples from learned angles are indistinguishable from solver angles.
- Topology-sensitive models improve angle quality; training cost is amortized over repeated solves.
From the original article
1 min read at arxiv.orgThe Quantum Approximate Optimization Algorithm (QAOA) and its variants are leading algorithms to help solve challenging combinatorial optimization problems. QAOA maps these problems to Ising Hamiltonians such that their ground state is the desired solution. Candidate solutions are then drawn from a variational quantum circuit.
By Johannes Jakubik · 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.