Speakers
Description
Variational quantum algorithms are promising for near-term quantum devices, but their optimization is hindered by barren plateaus and finite-shot noise. We introduce the Quantum Tilted Loss (QTL), a one-parameter nonlinear objective obtained by exponentially tilting measurement outcomes of an observable. The tilt selects the relevant spectral region: zero tilt recovers the expectation value, while negative tilt emphasizes low-energy outcomes. QTL preserves global minimizers, connects to CVaR and Gibbs/free-energy losses, and admits finite-shot estimators for diagonal Hamiltonians. Analytic benchmarks and QAOA MaxCut simulations show that moderate or scheduled tilts improve trainability, whereas excessive tilting becomes statistically costly.
| Topic | Quantum-enhanced robustness in machine learning models |
|---|---|
| Additional Topic | Quantum variational circuits and their applications |