Artificial intelligence (AI) is utilized in numerous industries, including those people might not immediately expect. One example is the way physicists are turning to AI to do things that aren’t otherwise possible.
Analyzing Huge Amounts of Data
Physicists who perform experiments with the Large Hadron Collider (LHC) know that they’re dealing with the largest particle accelerator in the world. Each test produces about a million gigabytes of data per second.
After only an hour, the LHC generates approximately as much data as Facebook handles in a year. That massive amount of information is too much for humans to process without help. Luckily, they get help from machine learning, a type of AI.
Before machine learning, physicists would build a specific computer algorithm to do a desired type of analysis, but they’d have to spend hours doing design and analysis work first. Machine learning saves them from completing those steps. It figures out analyses independently, leaving the physicists free to devote themselves to other tasks.
Hardware and software perform real-time evaluations to decide which data to keep and which to discard. Whether the machine learning algorithms sift through data or look at what’s left, physicists say they handle about 70 percent of the decisions associated with some LHC experiments.
In the summer of 2018, machine learning helped scientists detect an interaction between the Higgs boson and the top quark. As such, the physicists could confirm their strength.
The Higgs boson is particularly notable because researchers say it’s part of the Standard Model of physics. It contains the rules making up the universe’s fundamental building blocks.
Making the LHC More Capable
One of the primary reasons that physicists want to explore machine learning is because they think it could empower the LHC’s detector systems to process up to 20 times more data than they can currently.
The LHC records particle collisions with snapshots containing hundreds of thousands of pixels and up to 20 pairs of protons. Computers inside the LHC reconstruct the paths in real time, then move on to taking another picture.
Machine Learning May Reveal New Discoveries
Machine learning is also helping physicists learn more about how neutrinos change as they travel through Earth. They go through oscillations that turn them into different types.
As scientists study when and how that occurs, they believe they might find a new kind of neutrino that might be a particle of dark matter. Sensors notice the charged particles produced when the neutrons hit a detector, and machine learning identifies them.
Due to these uses for AI, it’s not surprising that Bill Gates refers to it as his “Holy Grail technology.” In the field of quantum physics, physicists use a computer program to design experiments they wouldn’t have thought of on their own. As such, they could realize new things that wouldn’t be evident without machine learning.
Juan Carrasquilla is a condensed matter expert who taught himself machine learning to propel his work. He believes machine learning could someday solve what he calls the “ultimate big data problem” for physicists: finding the wave pattern of a many-electron system. If that happened, the finding could lead to solved issues in materials science, plus multi-million-dollar pharmaceutical advancements.
Applying Deep Learning to Physics
Physicists also anticipate using a type of machine learning called deep learning to make even more progress in particle physics applications. It’s inspired by the neural networks in the human brain.
Until recently, it was extremely difficult to train extensive neural networks. However, algorithm and hardware improvements opened up new possibilities to solve long-standing particle physics problems.
Even though the marriage of deep learning and particle physics is still in its early stages, physicists are thinking positively about future possibilities. For example, they might use deep learning for computer vision tasks that recognize aspects of abstract images.
They might also use deep learning to identify features in particle jets, the narrow sprays of particles produced during some particle accelerator experiments. Researchers say it’s difficult to differentiate the jets into individualized tracks, but deep learning may help with that.
The “Black Box” Problem
Despite the fast progress in deep learning for particle physics, scientists remain cautious. They know machine learning algorithms are like “black boxes” that don’t give much information about how they reached particular conclusions.
When it’s not possible to fully understand the inner workings of an algorithm, physicists aim to perform cross-checks with real data to confirm the validity of an algorithm’s findings. They also work to ensure the algorithms don’t mistakenly throw away valuable data.
AI Is Pushing Physics Forward
This brief overview highlights why physicists see promise in the concept of applying AI to their work. The early improvements are notable, and researchers are working on ways to give machine learning a defining role in their attempts to find out more about the universe.