The field of application of machine learning in quantum computing received a boost amid new research that removes a potential obstacle to the practical implementation of quantum neural networks. Despite the theorists' current belief that training such a network would require an exponentially large dataset, the quantum No-Free-Lunch Theorem (NFL) developed by Los Alamos National Laboratory shows that quantum entanglement removes this exponential overhead.
“Our work proves that both big data and big entanglement are significant for quantum machine learning. Moreover, entanglement leads to scalability, saving us the complexity of exponentially increasing the size of the data needed for training.” says Andrew Sornborger, a computer scientist at the lab and co-author of a paper published Feb. 18 in Physical Review Letters. “This theorem gives us hope that quantum neural networks are on the path to accelerating quantum computing, where they will eventually outperform their current alternative running on the computers we are used to.” The classic No-Free-Lunch theorem states that any machine learning algorithm no worse or better than any other when their performance is averaged over all possible functions that bind data to labels. A direct consequence of this theorem, demonstrating the power of data influence in classical machine learning, is that the more data we have, the higher the average efficiency. Thus, data in ML is a kind of currency, in fact, limiting the effectiveness of its algorithms.
The new No-Free-Lunch theorem, developed by the Los Alamos Laboratory, shows that in quantum conditions, entanglement also acts as a currency, and one that can be exchanged for data, thereby reducing the need for them.
Using a Rigetti quantum computer to test a new theorem, the team realized the entanglement of a quantum dataset with a reference system.
“We have demonstrated on quantum hardware that we can definitely break the standard No-Free-Lunch theorem with entanglement. At the same time, our new formulation of it was confirmed in the course of the experiment.” said Kunal Sharma, one of the paper's authors. “Our theorem suggests that entanglement, along with big data, should be considered a valuable resource for quantum machine learning. says Patrick Coles, a Los Alamos laboratory physicist and lead author on the paper. “Classical neural networks depend only on big data.”
News translation: Entanglement unlocks scaling for quantum machine learning
bbabo.Net