Nuclear physics adopts machine learning

One specific activity that ML requires computers to complete is complex computations

To assist them save time and money, scientists have started utilizing new tools provided by machine learning. Nuclear physics has experienced a flurry of machine-learning initiatives going online during the past few years, and numerous papers have been published on the topic. "It was essential to record the work that had been completed. In order for people to better comprehend the variety of activities, we genuinely wish to raise awareness of the use of ML in nuclear physics, "Amber Boehnlein remarked.

Because it combines and analyses key work that has been done in the field thus far, Boehnlein believes that the essay will act as a roadmap for future studies as well as an instructional resource for people who are intrigued by the subject.

Boehnlein went to a workshop on artificial intelligence in March 2020 at Jefferson Lab. They were inspired to move forward after publishing a follow-up report. And including 15 other colleagues who covered all the subfields of nuclear physics, they decided to carry out a review of the current machine learning initiatives in nuclear physics.

As the authors point out, the first significant application of ML in nuclear physics took place in 1992, when computer experiments were used to look at nuclear features such as atomic weights. This work suggests the potential of machine learning, but its use in the sector has remained modest for more than two decades. Recent years have seen a change in that.

One specific activity that ML, which includes building models that can perform tasks without specific instruction, requires computers to complete is complex computations. It has become simpler for physicists to incorporate machine learning into their study because of recent advancements in computing capacity.

Machine learning is being applied in research at all energies and scales, from investigations into the elements of matter to investigations into the cyclic evolution of stars. Additionally, it is found in the four subspecialties of nuclear physics: accelerator science, theory, experiment, and operations.

The conception and execution of nuclear physics tests can both benefit from the use of ML models. Additionally, they can be used to aid in the analysis of the test data, which often exceeds petabytes.

Machine learning will speed up these processes, perhaps saving money on beamtime, computer utilization, and other experimental costs.

However, machine learning has experienced the most advancement in nuclear theory thus far. Nuclear theorist Nazarewicz is highly interested in this subject.

Using machine learning, he claims, theorists may make predictions, enhance and streamline their models, and comprehend the uncertainty surrounding those predictions. Additionally, it can be utilized to analyze phenomena that are way complex for experimental, such as neutron stars and supernova outbursts.

Boehnlein claims that Jefferson Lab theorists have started utilising these methods to investigate the structure of protons and neutrons. For instance, machine learning can aid in the information extraction process from the complex theory of quantum chromodynamics, which describes interactions between the gluons and quarks that comprise neutrons and protons.