• Topic Area
  • Subfield
  • Topic

Traina­bi­lity

Traina­bi­lity in quantum machine learning (QML) descri­bes how relia­bly and effici­ently model parame­ters can be optimi­zed. Many QML models rely on gradi­ents of a loss function, but these gradi­ents can become extre­mely small as circuits grow in number of qubits, making learning slow. This is usually called the “barren plateau” pheno­me­non. Traina­bi­lity also depends on circuit depth, parame­ter initia­liza­tion, choice of cost function, finite sampling noise, and hardware noise. Practi­cal strate­gies to improve traina­bi­lity include using problem-inspi­red or shallow circuit archi­tec­tures, local or struc­tu­red cost functions, parame­ter warm starts, layer­wise training, and noise-aware optimiza­tion. Asses­sing traina­bi­lity helps decide whether a QML model will scale beyond small demonstrations.

Topics in the Subfield Quantum machine learning:

Select another topic within the subfield Quantum machine learning

Genera­tive QML

In genera­tive quantum machine learning (GQML), the objec­tive is to learn a model from a given set of data points that can…

Super­vi­sed QML

Super­vi­sed machine learning (ML) is a funda­men­tal algorith­mic approach that enables…

Unsuper­vi­sed QML

Unsuper­vi­sed quantum machine learning explo­res how quantum compu­ters can identify hidden patterns in data without…

Kernel

Kernel methods map data with complex, nonlinear relati­onships into a high dimen­sio­nal feature space in order to…

QNNs

Quantum neural networks (QNNs) are machine learning models that…

QELM

Tradi­tio­nal approa­ches to super­vi­sed learning based on neural network archi­tec­tures (NNs) require, during the…

Partners