IOP Publishing’s Machine Learning series is the world’s first open-access journal series dedicated to the application and development of machine learning (ML) and artificial intelligence (AI) for the sciences.

Part of the series is Machine Learning: Science and Technology, launched in 2019, which bridges the application and advances in machine learning across the sciences. Machine Learning: Earth is dedicated to the application of ML and AI across all areas of Earth, environmental and climate sciences while Machine Learning: Health covers healthcare, medical, biological, clinical and health sciences and Machine Learning: Engineeringfocuses on applied AI and non-traditional machine learning to the most complex engineering challenges.

Here, the editors-in-chief (EiC) of the four journals discuss the growing importance of machine learning and their plans for the future.

Kyle Cranmer is a particle physicist and data scientist at the University of Wisconsin-Madison and is EiC of Machine Learning: Science and Technology (MLST). Pierre Gentine is a geophysicist at Columbia University and is EiC of Machine Learning: Earth. Jimeng Sun is a biophysicist at the University of Illinois at Urbana-Champaign and is EiC of Machine Learning: Health. Mechanical engineer Jay Lee is from the University of Maryland and is EiC of Machine Learning: Engineering.

What do you attribute to the huge growth over the past decade in research into and using machine learning?

Kyle Cranmer (KC): It is due to a convergence of multiple factors. The initial success of deep learning was driven largely by benchmark datasets, advances in computing with graphics processing units, and some clever algorithmic tricks. Since then, we’ve seen a huge investment in powerful, easy-to-use tools that have dramatically lowered the barrier to entry and driven extraordinary progress.

Pierre Gentine (PG): Machine learning has been transforming many fields of physics, as it can accelerate physics simulation, better handle diverse sources of data (multimodality), help us better predict.

Jimeng Sun (JS): Over the past decade, we have seen machine learning models consistently reach — and in some cases surpass — human-level performance on real-world tasks. This is not just in benchmark datasets, but in areas that directly impact operational efficiency and accuracy, such as medical imaging interpretation, clinical documentation, and speech recognition. Once ML proved it could perform reliably at human levels, many domains recognized its potential to transform labour-intensive processes.

Jay Lee (JL):  Traditionally, ML growth is based on the development of three elements: algorithms, big data, and computing.  The past decade’s growth in ML research is due to the perfect storm of abundant data, powerful computing, open tools, commercial incentives, and groundbreaking discoveries—all occurring in a highly interconnected global ecosystem.

What areas of machine learning excite you the most and why?

KC: The advances in generative AI and self-supervised learning are very exciting. By generative AI, I don’t mean Large Language Models — though those are exciting too — but probabilistic ML models that can be useful in a huge number of scientific applications. The advances in self-supervised learning also allows us to engage our imagination of the potential uses of ML beyond well-understood supervised learning tasks.

PG: I am very interested in the use of ML for climate simulations and fluid dynamics simulations.

JS: The emergence of agentic systems in healthcare — AI systems that can reason, plan, and interact with humans to accomplish complex goals. A compelling example is in clinical trial workflow optimization. An agentic AI could help coordinate protocol development, automatically identify eligible patients, monitor recruitment progress, and even suggest adaptive changes to trial design based on interim data. This isn’t about replacing human judgment — it’s about creating intelligent collaborators that amplify expertise, improve efficiency, and ultimately accelerate the path from research to patient benefit.

JL: One area is  generative and multimodal ML — integrating text, images, video, and more — are transforming human–AI interaction, robotics, and autonomous systems. Equally exciting is applying ML to nontraditional domains like semiconductor fabs, smart grids, and electric vehicles, where complex engineering systems demand new kinds of intelligence.

What vision do you have for your journal in the coming years?

KC: The need for a venue to propagate advances in AI/ML in the sciences is clear. The large AI conferences are under stress, and their review system is designed to be a filter not a mechanism to ensure quality, improve clarity and disseminate progress. The large AI conferences also aren’t very welcoming to user-inspired research, often casting that work as purely applied. Similarly, innovation in AI/ML often takes a back seat in physics journals, which slows the propagation of those ideas to other fields. My vision for MLST is to fill this gap and nurture the community that embraces AI/ML research inspired by the physical sciences.

PG: I hope we can demonstrate that machine learning is more than a nice tool but that it can play a fundamental role in physics and Earth sciences, especially when it comes to better simulating and understanding the world.

JS: I see Machine Learning: Health becoming the premier venue for rigorous ML–health research — a place where technical novelty and genuine clinical impact go hand in hand. We want to publish work that not only advances algorithms but also demonstrates clear value in improving health outcomes and healthcare delivery. Equally important, we aim to champion open and reproducible science. That means encouraging authors to share code, data, and benchmarks whenever possible, and setting high standards for transparency in methods and reporting. By doing so, we can accelerate the pace of discovery, foster trust in AI systems, and ensure that our field’s breakthroughs are accessible to — and verifiable by — the global community.

JL:  Machine Learning: Engineering envisions becoming the global platform where ML meets engineering. By fostering collaboration, ensuring rigour and interpretability, and focusing on real-world impact, we aim to redefine how AI addresses humanity’s most complex engineering challenges.

The post Physicists discuss the future of machine learning and artificial intelligence appeared first on Physics World.

China has delayed the return of a crewed mission to the country’s space station over fears that the astronaut’s spacecraft has been struck by space debris. The craft was supposed to return to Earth on 5 November but the China Manned Space Agency says it will now carry out an impact analysis and risk assessment before making any further decisions about when the astronauts will return.

The Shenzhou programme involves taking astronauts to and from China’s Tiangong space station, which was constructed in 2022, for six-month stays.

Shenzhou-20, carrying three crew, launched on 24 April from Jiuquan Satellite Launch Center on board a Long March 2F rocket. Once docked with Tiangong the three-member crew of Shenzhou-19 began handing over control of the station to the crew of Shenzhou-20 before they returned to Earth on 30 April.

The three-member crew of Shenzhou-21 launched on 31 October and underwent the same hand-over process with the crew of Shenzhou-20 before they were set to return to Earth on Wednesday.

Yet pre-operation checks revealed that the craft had been hit by “a small piece of debris” with the location and scale of the damage to Shenzhou-20 having not been released.

If the craft is deemed unsafe following the assessment, it is possible that the crew of Shenzhou-20 will return to Earth aboard Shenzhou-21. Another option is to launch a back-up Shenzhou spacecraft, which remains on stand-by and could be launched within eight days.

Space debris is of increasing concern and this marks the first time that a crewed craft has been delayed due to a potential space debris impact. In 2021, for example, China noted that Tiangong had to perform two emergency avoidance manoeuvres to avoid fragments produced by Starlink satellites that were launched by SpaceX.

• For more on the impact of space debris, sign-up for a Physics World Live event on “Space junk – and how to solve it” on 10 November at 9 p.m. GMT.

The post China’s Shenzhou-20 crewed spacecraft return delayed by space debris impact appeared first on Physics World.

It’s Halloween today and so what better time than to bring you a couple of spooky stories from the world of physics.

First up is researchers at the University of Georgia in the US who have confirmed that six different species of bats found in North America emit a ghoulish green light when exposed to ultraviolet light.

The researchers examined 60 specimens from the Georgia Museum of Natural History and exposed the bats to UV light.

They found that the wings and hind limbs of six species – big brown bats, eastern red bats, Seminole bats, southeastern myotis, grey bats and the Brazilian free-tailed bat – gave off photoluminescence with the resulting glow being a shade of green.

While previous research found that some mammals, like pocket gophers, also emit a glow under ultraviolet light, this was the first discovery of such a phenomenon for bats located in North America.

The colour and location of the glow on the winged mammals suggest it is not down to genetics or camouflage and as it is the same between sexes it is probably not used to attract mates.

“It may not seem like this has a whole lot of consequence, but we’re trying to understand why these animals glow,” notes wildlife biologist Steven Castleberry from the University of Georgia.

Given that many bats can see the wavelengths emitted, one option is that the glow may be an inherited trait used for communication.

“The data suggests that all these species of bats got it from a common ancestor. They didn’t come about this independently,” adds Castleberry. “It may be an artifact now, since maybe glowing served a function somewhere in the evolutionary past, and it doesn’t anymore.”

Thread lightly

In other frightful news, spider webs are a classic Halloween decoration and while the real things are marvels of bioengineering, there is still more to understand about these sticky structures.

Many spider species build spiral wheel-shaped webs – orb webs – to capture prey, and some incorporate so-called “stabilimenta” into their web structure. These “extra touches” look like zig-zagging threads that span the gap between two adjacent “spokes,” or threads arranged in a circular “platform” around the web’s centre.

The purpose of stabilimenta is unknown and proposed functions include as a deterrence for predatory wasps or birds.

Yet Gabriele Greco of the Swedish University of Agricultural Sciences and colleagues suggest such structures might instead influence the propagation of web vibrations triggered by the impact of captured prey.

Greco and colleagues observed different stabilimentum geometries that were constructed by wasp spiders, Argiope bruennichi. The researchers then performed numerical simulations to explore how stabilimenta affect prey impact vibrations.

For waves generated at angles perpendicular to the threads spiralling out from the web centre, stabilimenta caused negligible delays in wave propagation.

However, for waves generated in the same direction as the spiral threads, vibrations in webs with stabilimenta propagated to a greater number of potential detection points across the web – where a spider might sense them – than in webs without stabilimenta.

This suggests that stabilimenta may boost a spider’s ability to pinpoint the location of unsuspecting prey caught in its web.

Spooky.

The post Spooky physics: from glowing green bats to vibrating spider webs appeared first on Physics World.