Luisa Crawford
Sep 30, 2024 17:41
NVIDIA collaborates with the College of Toronto and Saint Jude Youngsters’s Analysis Hospital to combine GPT fashions into quantum algorithm design.
Synthetic intelligence (AI) methods, significantly giant language fashions (LLMs), are revolutionizing varied scientific fields, and quantum computing isn’t any exception. In line with the NVIDIA Technical Weblog, a collaboration between NVIDIA, the College of Toronto, and Saint Jude Youngsters’s Analysis Hospital is leveraging generative pre-trained transformers (GPTs) to design new quantum algorithms, together with the Generative Quantum Eigensolver (GQE) method.
Revolutionary GQE Approach
The GQE method represents a big development within the area of AI for Quantum. Developed utilizing the NVIDIA CUDA-Q platform, GQE is the primary methodology permitting customers to make use of their very own GPT fashions to create advanced quantum circuits. The CUDA-Q platform, with its emphasis on accelerated quantum supercomputing, gives a hybrid computing surroundings ideally suited to the GQE method.
Alan Aspuru-Guzik, a co-author of the GQE methodology, highlights the scalable nature of the CUDA-Q platform, which integrates CPUs, GPUs, and QPUs for coaching and utilizing GPT fashions in quantum computing.
Understanding GQE by means of LLM Analogy
Typical LLMs, which perceive and generate textual content by studying from giant datasets, present a helpful analogy for comprehending GQE. Whereas LLMs cope with phrases, GQE offers with quantum circuit operations. GQE makes use of a transformer mannequin to generate sequences of unitary operations, which outline quantum circuits. The coaching includes minimizing a price operate evaluated by computing expectation values of beforehand generated circuits.
Developments in NISQ Period Algorithms
Within the noisy, small-scale quantum (NISQ) period, quantum algorithms face a number of {hardware} constraints. Hybrid quantum-classical algorithms just like the Variational Quantum Eigensolver (VQE) goal to beat these limitations. The GQE methodology, nonetheless, is the primary to leverage AI to speed up NISQ functions, optimizing parameters classically inside the GPT mannequin.
GQE extends NISQ algorithms by:
Constructing quantum circuits with out quantum variational parameters.
Bettering quantum useful resource effectivity by changing quantum gradient analysis with sampling and backpropagation.
Permitting customization to include area information or goal functions outdoors of chemistry.
Enabling pretraining to remove the necessity for added quantum circuit evaluations.
Outcomes and Future Prospects
For its inaugural utility, the GQE mannequin, impressed by GPT-2 and known as GPT-QE, was used to estimate floor state energies of small molecules. The mannequin demonstrated important parallelization capabilities, lowering coaching instances from 173 hours on a single NVIDIA H100 GPU to three.5 hours throughout 48 H100 GPUs.
Future analysis goals to discover completely different operator swimming pools for GQE and optimum coaching methods, together with pretraining. This course of makes use of current datasets to boost the effectivity of transformer coaching or help in its convergence.
Past quantum chemistry, NVIDIA and Los Alamos Nationwide Lab are investigating the applying of GQE to geometric quantum machine studying, showcasing the broad potential of this revolutionary method.
For extra details about this collaboration and the GQE code, go to the NVIDIA Technical Weblog.
Picture supply: Shutterstock
