Q&A with Anatole von Lilienfeld, chief editor of APS’ newest open access journal

On Nov. 19, APS debuted PRX Intelligence, the newest journal in the highly selective Physical Review portfolio. Covering machine learning and artificial intelligence in scientific research, this open access journal will cater to physical scientists who rely on data-intensive approaches.

Leading the effort is the journal’s chief editor, Anatole von Lilienfeld, a professor of chemistry, materials, and physics at the University of Toronto. Von Lilienfeld is also the Clark Chair of Advanced Materials and CIFAR AI chair at the Vector Institute for Artificial Intelligence.

“I’m very attracted to the open science culture of APS,” says von Lilienfeld. “I feel honored to be given this opportunity.”

After earning his doctorate in 2005 from the Swiss Federal Institute of Technology in Lausanne, von Lilienfeld held research positions at New York University, the Max-Planck Institute for Polymer Research in Germany, and Sandia and Argonne National Laboratories. He also held faculty positions at universities in Belgium, Switzerland, and Austria.

Von Lilienfeld has focused on developing methods to study materials and chemical compound spaces, relying on machine learning, quantum and statistical mechanics, and high-performance computing. He also has deep editorial experience, having served as the editor in chief of Machine Learning: Science and Technology; associate editor of Science Advances, Journal of the American Chemical Society, and Journal of Chemical Theory and Computation; and editorial board member of Scientific Data.

APS News spoke with von Lilienfeld to learn more about what the APS community can expect from PRX Intelligence.

This interview has been edited for length and clarity

What makes the timing right for PRX Intelligence?

Physics started like all the sciences — with experiments. Then theory was developed, and we arrived at theoretical frameworks with some equations that were too difficult to solve.

Since the digital revolution, we’ve used computers to numerically solve approximations to these equations, and to accelerate the experiments. We’ve enjoyed more and cheaper access to computation, thanks to Moore’s law. And by combining these three pillars of science, we’ve gotten very large datasets, which makes more of the field amenable to statistical learning.

Now, we’ve seen the emergence of AI, a fourth pillar of science. So a new journal dedicated to AI and physics is timely.

What attracted you to the chief editor role?

My own research has benefited from physics, which I’ve incorporated into my work developing machine learning models. I think this lesson holds in general — that there’s an important role that physics can play in developing new and improved machine learning approaches.

I’m very excited about these opportunities. I felt that the ideal publisher for this type of work is APS, and the community of APS is the perfect audience for it.

What’s your vision for PRX Intelligence?

AI is a common theme in research right now, but in a lot of work being submitted to journals, it’s not always clear what advancement an insight could lead to.

APS has a strong reputation for high-quality content. We want to preserve, or even raise, the standards of APS journals by trying to identify studies that truly represent a major new insight or advancement in the field, and to relate those advances to physics.

PRX Intelligence is also run by scientists for scientists. We want editors and referees to do full justice to the authors’ work. All our editors and editorial board members are active researchers, and we’ll match submitted manuscripts with the most competent referees.

From your own work, what challenge might machine learning research help resolve?

In the physical sciences, we have always struggled to establish a robust understanding to make more of our approaches predictive. The hope is, given enough data and machine learning models for training, we will be able to improve the situation, which could benefit not only our understanding but also the design and discovery of new materials, which could be useful for renewable energy or new chip designs.

Another challenge from my field is chemical and materials synthesis. Even if you’ve identified a material that you think has all the properties and behaviors you’d like to see, its synthesis is still a challenge. We hope we can make progress using new tools to design synthesis routes.

These are typical problems for atomic and materials sciences — my background — but other fields will also benefit, because statistical methods manifest themselves in a variety of approaches. They include supervised learning, unsupervised learning, generative learning, AI — a lot of new tools that have become ever more useful.

Describe the primary audience for PRX Intelligence.

I view my personal research interests in the physics of the chemical and materials sciences as representative of the kinds of readers we’d like to address. But we are also addressing researchers in other domains of physics who have connected theory to simulation, AI, and experiments. We would especially like to involve those who combine data with physics and statistical insights to train more predictive models.

What new or surprising features might we see in the journal?

From my own work, where I’ve contributed data sets — such as QM7, QM9, or VQM24 — that have proven important for the field, it’s shown me that a new article form such as a ‘data paper’ could be valuable. There’s a need for a journal that can offer that in the APS family of journals. We will also publish new code to facilitate the dissemination of methods and the reproducibility of results. These are just a few of our ideas to make this journal useful for authors and readers.

What’s been challenging about the journal process for you?

Finding the right people. This is a journal by scientists, for scientists, so we’ve tried to cover all the domains of physics that could benefit from machine learning, without singling out a particular trend. We want to be an open-minded journal. So far, the leadership team has representatives from condensed matter physics, particle physics, nuclear physics, and astrophysics, as well as a member who’s an experimentalist.

What else would you like the APS community to know about PRX Intelligence?

The peer review system has faced criticism, and it’s especially problematic when a referee is asked to make the call on whether a paper is appropriate for the journal, and the referee says, ‘It’s a good paper but not the right fit for the journal.’

For PRX Intelligence, we’re working to avoid this issue by no longer asking referees if a paper is appropriate for the journal or not. Our editors will first make the call on whether the paper is appropriate, and then the referees will make recommendations to improve it.

The focus will be on improving the quality of the content, rather than fighting to get published. It’s a cultural change we are excited about.

Why does it matter that the journal is open access?

My personal philosophy has always been to make heavy use of arXiv and open source — to not hide research behind paywalls. This is particularly important for scientists in developing countries where academic libraries cannot necessarily afford subscription fees.

We want people from around the globe to be able to access the content, because we believe that sharing science openly helps humanity as a whole. This is something that the open access model enables.