Quantum Computers Could Process Massive AI Datasets With Less Memory, Study Finds
New method prepares quantum states during processing instead of loading full datasets, potentially allowing systems with 60 logical qubits to outperform classical computers
Quick Look
- Researchers from Caltech, Google Quantum AI, Oratomic, and MIT have developed a method allowing quantum computers to process massive AI training datasets with reduced memory requirements.
- The approach prepares quantum states during processing rather than loading entire datasets into quantum memory first.
- A system with about 300 logical qubits could outperform classical computers on certain tasks, while roughly 60 logical qubits could begin outperforming classical systems on some AI data-processing tasks—though such systems do not yet exist.
AI-generated summary
Quantum computers may eventually help process some of the massive datasets used to train artificial intelligence, according to a report by New Scientist. Drawing from an earlier study by Caltech, Google Quantum AI, quantum computing startup Oratomic, and MIT, researchers say one challenge has been getting large datasets—often measured in terabytes or petabytes—into a quantum computer. To use quantum effects, data must be converted into a quantum state, and preparing those states has traditionally required significant quantum memory.
“Machine learning is really utilized everywhere in science and technology, and also everyday life. In a world where we can build this [quantum computing] architecture, I feel like it can be applied whenever there’s massive datasets available,” Hsin-Yuan Huang, CTO at Oratomic, said in a statement.
The study proposes that, rather than requiring the full dataset to be loaded into quantum memory first, the new method prepares the necessary quantum states during processing, reducing the memory burden. The researchers say this could allow quantum effects such as superposition to be used without extremely large storage systems.
The researchers say the approach could also allow quantum computers to process large datasets while using less memory than conventional systems, suggesting that a machine with about 300 logical qubits—error-corrected quantum bits that can reliably perform calculations—could outperform classical computers on certain tasks. Such a system does not yet exist; however, the researchers estimate that a quantum computer with roughly 60 logical qubits could begin outperforming classical systems on some data-processing tasks used in artificial intelligence, highlighting how advances in quantum computing could threaten fields such as cryptography and blockchain.
“People are used to quantum computers always being 10 years away,” Oratomic co-founder and CEO Dolev Bluvstein previously told Decrypt. “But when you look at where we were a little over ten years ago, the best estimates of what would be required for Shor’s algorithm were one billion qubits at a time when the best systems we had in the lab were roughly five qubits.”
Still, researchers say the connection between artificial intelligence and quantum computing is growing closer, as AI tools help scientists analyze and model complex quantum systems that would otherwise be difficult to simulate, accelerating work on quantum hardware and applications.
“The quantum machine is a very powerful device, but you do need to first feed it,” Professor of Computational Physics at ETH Zurich in Switzerland, Adrián Pérez-Salinas, said in a statement. “This study talks about feeding and how it’s enough to load [data] bit by bit, without overfeeding the beast.”
