Connecting ansatz expressibility to gradient magnitudes and barren plateaus

  title={Connecting ansatz expressibility to gradient magnitudes and barren plateaus},
  author={Zoe Holmes and Kunal Sharma and Mar{\'i}a Cerezo and Patrick J. Coles},
Parameterized quantum circuits serve as ansätze for solving variational problems and provide a flexible paradigm for programming near-term quantum computers. Ideally, such ansätze should be highly expressive so that a close approximation of the desired solution can be accessed. On the other hand, the ansatz must also have sufficiently large gradients to allow for training. Here, we derive a fundamental relationship between these two essential properties: expressibility and trainability. This is… 

Figures from this paper

Avoiding barren plateaus via transferability of smooth solutions in a Hamiltonian variational ansatz

A large ongoing research effort focuses on Variational Quantum Algorithms (VQAs), representing leading candidates to achieve computational speed-ups on current quantum devices. The scalability of VQAs

Diagnosing Barren Plateaus with Tools from Quantum Optimal Control

Variational Quantum Algorithms (VQAs) have received considerable attention due to their potential for achieving near-term quantum advantage. However, more work is needed to understand their

Higher order derivatives of quantum neural networks with barren plateaus

Novel, general formulas are proved that can be used to analytically evaluate any high-order partial derivative on quantum hardware and show the exponential suppression of higher order derivatives in a barren plateau phenomenon.

Classical Splitting of Parametrized Quantum Circuits

Barren plateaus appear to be a major obstacle to using variational quantum algorithms to simulate large-scale quantum systems or replace traditional machine learning algorithms. They can be caused by

Reducing Barren Plateaus in Quantum Algorithm Protocols

An approach to quantum algorithm optimization that is based on trainable Fourier coef-ficients of Hamiltonian system parameters and the ansatz, which applies to the extension of discrete quantum variational algorithms to analogue quantum optimal control schemes and is non-local in time.

A perturbative gadget for delaying the onset of barren plateaus in variational quantum algorithms

This work introduces a novel perturbative gadget, tailored to variational quantum algorithms, that can be used to delay the onset of barren plateaus and provides guarantees on the closeness of global minima and prove that the local cost function defined by the three-body Hamiltonian exhibits non-vanishing gradients.

Surviving The Barren Plateau in Variational Quantum Circuits with Bayesian Learning Initialization

The fast-and-slow algorithm is introduced, which uses Bayesian Learning to identify a promising region in parameter space and is used to initialize a fast local optimizer to find the global optimum point efficiently.

The quantum cost function concentration dependency on the parametrization expressivity

The analytically show that the more expressive the parametrization is, the more the cost function will tend to concentrate around a value that depends both on the chosen observable and on the number of qubits used.

Equivalence of quantum barren plateaus to cost concentration and narrow gorges

This work analytically proves the connection between three different landscape features that have been observed for PQCs: exponentially vanishing gradients, exponential cost concentration about the mean, and the exponential narrowness of minina.

CAFQA: Clifford Ansatz For Quantum Accuracy

CAFQA, a Clifford ansatz for quantum accuracy, is proposed and shown to be a highly accurate ansatz initialization for variational algorithms, highlighting the potential for quantum-inspired classical techniques as both an alternative as a supporting methodology for VQAs in the NISQ era and beyond.



Equivalence of quantum barren plateaus to cost concentration and narrow gorges

This work analytically proves the connection between three different landscape features that have been observed for PQCs: exponentially vanishing gradients, exponential cost concentration about the mean, and the exponential narrowness of minina.

Entanglement Induced Barren Plateaus

It is shown that for any bounded objective function on the visible layers, the Lipshitz constants of the expectation value of that objective function will scale inversely with the dimension of the hidden-subsystem with high probability, and how this can cause both gradient descent and gradient-free methods to fail.

On barren plateaus and cost function locality in variational quantum algorithms

A lower bound on the variance of the gradient is derived, which depends mainly on the width of the circuit causal cone of each term in the Pauli decomposition of the cost function.

Entanglement devised barren plateau mitigation

This work defines barren plateaus in terms of random entanglement and proposes and demonstrates a number of barren plateau ameliorating techniques, including initial partitioning of cost function and non-cost function registers, meta-learning of lowentanglement circuit initializations, selective inter-register interaction, entanglements regularization, and rotation into preferred cost function eigenbases.

Noise-induced barren plateaus in variational quantum algorithms

This work rigorously proves a serious limitation for noisy VQAs, in that the noise causes the training landscape to have a barren plateau, and proves that the gradient vanishes exponentially in the number of qubits n if the depth of the ansatz grows linearly with n.

Expressibility and trainability of parametrized analog quantum systems for machine learning applications

This work investigates how the interplay between external driving and disorder in the system dictates the trainability and expressibility of interacting quantum systems, and devise a protocol using quenched MBL dynamics which allows accurate trainability while keeping the overall dynamics in the quantum supremacy regime.

Effect of barren plateaus on gradient-free optimization

It is shown that gradient-free optimizers do not solve the barren plateau problem, and the main result proves that cost function differences, which are the basis for making decisions in a gradient- free optimization, are exponentially suppressed in a barren plateau.

From the Quantum Approximate Optimization Algorithm to a Quantum Alternating Operator Ansatz

The essence of this extension, the quantum alternating operator ansatz, is the consideration of general parameterized families of unitaries rather than only those corresponding to the time evolution under a fixed local Hamiltonian for a time specified by the parameter.

Absence of Barren Plateaus in Quantum Convolutional Neural Networks

This work rigorously analyze the gradient scaling for the parameters in the QCNN architecture and finds that the variance of the gradient vanishes no faster than polynomially, implying that QCNNs do not exhibit barren plateaus.

The theory of variational hybrid quantum-classical algorithms

This work develops a variational adiabatic ansatz and explores unitary coupled cluster where it is shown how the use of modern derivative free optimization techniques can offer dramatic computational savings of up to three orders of magnitude over previously used optimization techniques.