A group of international researchers have discovered a significant obstacle preventing training in quantum machine learning: excess quantum entanglement.
Quantum machine learning studies the benefits of quantum computers for artificial intelligence (AI). The hope is that in the future, quantum neural networks will be able to combine the strengths of quantum computing and traditional neural networks, however, recent theoretical research indicates potential difficulties.
Machine learning requires algorithms to learn from data at a stage known as training. During the training, the algorithm gradually improves in the assigned task. However, it is mathematically proven that a large class of quantum algorithms see only negligible improvement due to a phenomenon known as the arid plateau, first reported by a team at Google in 2018. Experience of an arid plateau may prevent learning of the quantum algorithm.
Theoretical research, published in PRX Quantum, further investigates the causes of sterile trays with a new focus on the impact of excessive entanglement. The entanglement of qubits or quantum bits is a quantum effect that allows the exponential acceleration of quantum computing power.
Although entanglement is necessary for quantum acceleration, the research underscores the need for careful design of which qubits should be entangled and to what extent,” says research co-author Dr Maria Kieferova, , researcher at the ARC Center for Quantum Computing and Communication. Technology-based. at Sydney University of Technology.
This contradicts the popular idea that more quantum entanglement provides faster accelerations.
We have shown that excessive entanglement between the output qubits, or visible units, and the rest of the quantum neural network hinders the learning process and that large amounts of entanglement can be catastrophic for the model.” , says lead author Dr. Carlos Ortiz Marrero, who is currently a research assistant professor at North Carolina State University.
This result teaches us what quantum neural network structures we must avoid for successful algorithms.
Although research has shown that a range of simple translations of classic machine learning models to the quantum domain is not beneficial, there is a way forward,” says Dr Ortiz Marrero.
By limiting the depth and connectivity of the network, we may be able to avoid regimes into which quantum machine learning algorithms cannot be trained.
This can be achieved by precisely and purposefully implementing entanglement in quantum machine learning models. Wiebe, University of Toronto.
