Exploring The Power Of Data In Quantum Machine Learning

0
4988
Source:https://arxiv.org/pdf/2011.01938.pdf

Quantum computers have the capability to develop quantum machine learning algorithms. These algorithms can achieve better performance for modeling quantum-mechanical systems such as molecules, catalysts, or high-temperature superconductors. Since it is difficult for classical computers to handle the interference of the exponentially evolving states in the quantum world, quantum computers are expected to have an advantage in quantum originated-machine learning problems. The quantum advantage extends to machine learning problems in the classical domain, for example, computer vision or natural language processing.

TensorFlow image
https://blog.tensorflow.org/2020/11/characterizing-quantum-advantage-in.html

Fig1:Classical computers can solve a problem better with the help of data obtained in nature (e.g., physical experiments). However, an even broader class of problems can be solved using quantum computers.

Some questions regarding quantum models’ liability are addressed in this recent paper by Google Research and Caltech researchers. It describes the development of a mathematical framework to compare classical modeling approaches (neural networks, tree-based models, etc.) and quantum modeling approaches to understand potential advantages in making more accurate predictions. This framework applies to both types of data, i.e., from the classical world (MNIST, product reviews, etc.) and a quantum experiment (chemical reaction, etc.).

AdvertisementCoursera Plus banner featuring Johns Hopkins University, Google, and University of Michigan courses highlighting data science career-advancing content

The experiments show that by taking data from physical experiments obtained in nature, including experiments for exploring new catalysts, superconductors, or pharmaceuticals, classical ML models may obtain some degree of generalization beyond the training data. 

This paper also includes a quantitative comparison between the number of samples needed by classical and quantum models to make appropriate predictions. It also consists of a constructive approach to generate datasets that are hard to learn with specific classical models.

TensorFlow study graphic
https://blog.tensorflow.org/2020/11/characterizing-quantum-advantage-in.html

Fig 2: shows one of the data sets for which all attempted ML methods fail, but quantum methods do not. These trends are examined with the help of TensorFlow Quantum, an open-source library for quantum ML.  

The framework and tools are expected to allow the TensorFlow community to explore precursors of datasets that require quantum computers to make accurate predictions. 

Source: https://blog.tensorflow.org/2020/11/characterizing-quantum-advantage-in.html

Paper: https://arxiv.org/pdf/2011.01938.pdf

Advertisement

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.