There’s nice deal of debate within the quantum computing (QC) world about operating AI on quantum {hardware} and the event of so-called QuantumAI. The promise appears excessive, however so do the obstacles, such because the excessive quantity of information utilized in AI mannequin coaching.
Trapped ion quantum laptop specialist, Quantinuum, right this moment posted a weblog, “Quantum Computer systems will Make AI Higher,” that briefly critiques progress on adapting AI strategies and suggests larger adjustments are forward.
“Coaching fashions like ChatGPT requires processing huge datasets with billions, even trillions, of parameters. This calls for immense computational energy, usually unfold throughout hundreds of GPUs or specialised {hardware} accelerators. The environmental value is staggering—merely coaching GPT-3, as an example, consumed practically 1,300 megawatt-hours of electrical energy, equal to the annual power use of 130 common U.S. houses…
“Regardless of these challenges, the push to develop ever-larger fashions exhibits no indicators of slowing down.
“Enter quantum computing. Quantum expertise provides a extra sustainable, environment friendly, and high-performance answer—one that can essentially reshape AI, dramatically decreasing prices and growing scalability, whereas overcoming the restrictions of right this moment’s classical programs.”
Whereas the weblog is promotional for Quantinuum’s AI work, emphasizing its deep experience in that space, it’s additionally a very good learn for these unfamiliar with how AI instruments and strategies are being tailored to run on quantum computer systems.
The weblog’s principal focus is on changing pure language processing strategies to be used on quantum {hardware}; it’s an space on which Quantinuum argues nice progress has been made. Quantinuum showcases its work changing recurrent neural networks into parameterized quantum circuits (PQS):
“In a latest experiment, the workforce used their quantum RNN to carry out a regular NLP process: classifying film critiques from Rotten Tomatoes as constructive or damaging. Remarkably, the quantum RNN carried out in addition to classical RNNs, GRUs, and LSTMs, utilizing solely 4 qubits. This result’s notable for 2 causes: it exhibits that quantum fashions can obtain aggressive efficiency utilizing a a lot smaller vector area, and it demonstrates the potential for vital power financial savings in the way forward for AI,” report the weblog writers.
Inside a Quantinuum entice. (Supply: Quantinuum)
“In the same experiment, our workforce partnered with Amgen to make use of PQCs for peptide classification, which is a regular process in computational biology. Engaged on the Quantinuum System Mannequin H1, the joint workforce carried out sequence classification (used within the design of therapeutic proteins), and so they discovered aggressive efficiency with classical baselines of the same scale. This work was our first proof-of-concept utility of near-term quantum computing to a process crucial to the design of therapeutic proteins, and helped us to elucidate the route towards larger-scale purposes on this and associated fields, according to our {hardware} growth roadmap.”
Work on transformers and tensor networks are additionally briefly reviewed.
A strong extra public dialogue of how you can mix AI and quantum computing is admittedly simply beginning to emerge. Clearly the quantum world doesn’t need to be overwhelmed or slowed by the AI revolution. Making sense of how you can mix the 2 is prone to take awhile. That mentioned, Quantinuum, like many within the quantum growth group, consider QuantumAI will grow to be a giant ($) factor.
The weblog concludes, “As quantum computing {hardware} continues to enhance, quantum AI fashions could more and more complement and even substitute classical programs. By leveraging quantum superposition, entanglement, and interference, these fashions supply the potential for vital reductions in each computational value and power consumption. With fewer parameters required, quantum fashions may make AI extra sustainable, tackling one of many greatest challenges dealing with the business right this moment.”
Hyperlink to Quantinuum weblog, https://www.quantinuum.com/weblog/quantum-computers-will-make-ai-better.
This text first appeared on sister website HPCwire.