Mhiri, Mme Hela (2024) Local characterization of Variational Quantum Algorithms landscape PFE - Project Graduation, ENSTA.
![]()
| PDF 889Kb |
Abstract
Variational quantum algorithms (VQAs) are proposed as an eligible candidate for the programming of near-term quantum computers. However, the average behavior of such models shows that they do suffer from exponentially vanishing gradients in the number of qubits, rendering the model inefficient and non scalable. This has motivated the search for strategies to mitigate the effect of this exponential concentration. In this work, we focus on the local characterization of the loss function variance and show that for a restricted region that depends on the number of independent parameters and the circuit generators, the loss function variance decays only polynomially in the number of qubits. Our result can be then used as a building block to prove theoretical guarantees of non random initialization strategies, paving the way to efficiently trainable VQAs.
Item Type: | Thesis (PFE - Project Graduation) |
---|---|
Uncontrolled Keywords: | Variational Quantum Algorithms, Barren Plateaus, warm starts, Narrow gorge, learning theory |
Subjects: | Information and Communication Sciences and Technologies Mathematics and Applications Physics, Optics |
ID Code: | 10386 |
Deposited By: | Hela MHIRI |
Deposited On: | 04 oct. 2024 17:50 |
Dernière modification: | 04 oct. 2024 17:50 |
Repository Staff Only: item control page