Computers Math

In the future, artificial intelligence will play an important role in medicine. In diagnostics, successful tests have already been performed: for example, the computer can learn to categorise images with great accuracy according to whether they show pathological changes or not. However, it is more difficult to train an artificial intelligence to examine the time-varying conditions of patients and to calculate treatment suggestions — this is precisely what has now been achieved at TU Wien in cooperation with the Medical University of Vienna.
With the help of extensive data from intensive care units of various hospitals, an artificial intelligence was developed that provides suggestions for the treatment of people who require intensive care due to sepsis. Analyses show that artificial intelligence already surpasses the quality of human decisions. However, it is now important to also discuss the legal aspects of such methods.
Making optimal use of existing data
“In an intensive care unit, a lot of different data is collected around the clock. The patients are constantly monitored medically. We wanted to investigate whether these data could be used even better than before,” says Prof. Clemens Heitzinger from the Institute for Analysis and Scientific Computing at TU Wien (Vienna). He is also Co-Director of the cross-faculty “Center for Artificial Intelligence and Machine Learning” (CAIML) at TU Wien.
Medical staff make their decisions on the basis of well-founded rules. Most of the time, they know very well which parameters they have to take into account in order to provide the best care. However, the computer can easily take many more parameters than a human into account — and in some cases this can lead to even better decisions.
The computer as planning agent
“In our project, we used a form of machine learning called reinforcement learning,” says Clemens Heitzinger. “This is not just about simple categorisation — for example, separating a large number of images into those that show a tumour and those that do not — but about a temporally changing progression, about the development that a certain patient is likely to go through. Mathematically, this is something quite different. There has been little research in this regard in the medical field.”

The computer becomes an agent that makes its own decisions: if the patient is well, the computer is “rewarded.” If the condition deteriorates or death occurs, the computer is “punished.” The computer programme has the task of maximising its virtual “reward” by taking actions. In this way, extensive medical data can be used to automatically determine a strategy which achieves a particularly high probability of success.
Already better than a human
“Sepsis is one of the most common causes of death in intensive care medicine and poses an enormous challenge for doctors and hospitals, as early detection and treatment is crucial for patient survival,” says Prof. Oliver Kimberger from the Medical University of Vienna. “So far, there have been few medical breakthroughs in this field, which makes the search for new treatments and approaches all the more urgent. For this reason, it is particularly interesting to investigate the extent to which artificial intelligence can contribute to improve medical care here. Using machine learning models and other AI technologies are an opportunity to improve the diagnosis and treatment of sepsis, ultimately increasing the chances of patient survival.”
Analysis shows that AI capabilities are already outperforming humans: “Cure rates are now higher with an AI strategy than with purely human decisions. In one of our studies, the cure rate in terms of 90-day mortality was increased by about 3% to about 88%,” says Clemens Heitzinger.
Of course, this does not mean that one should leave medical decisions in an intensive care unit to the computer alone. But the artificial intelligence may run along as an additional device at the bedside — and the medical staff can consult it and compare their own assessment with the artificial intelligence’s suggestions. Such artificial intelligences can also be highly useful in education.
Discussion about legal issues is necessary
“However, this raises important questions, especially legal ones,” says Clemens Heitzinger. “One probably thinks of the question who will be held liable for any mistakes made by the artificial intelligence first. But there is also the converse problem: what if the artificial intelligence had made the right decision, but the human chose a different treatment option and the patient suffered harm as a result?” Does the doctor then face the accusation that it would have been better to trust the artificial intelligence because it comes with a huge wealth of experience? Or should it be the human’s right to ignore the computer’s advice at all times?
“The research project shows: artificial intelligence can already be used successfully in clinical practice with today’s technology — but a discussion about the social framework and clear legal rules are still urgently needed,” Clemens Heitzinger is convinced. More

Cornell University researchers have developed a robot, called ReMotion, that occupies physical space on a remote user’s behalf, automatically mirroring the user’s movements in real time and conveying key body language that is lost in standard virtual environments.
“Pointing gestures, the perception of another’s gaze, intuitively knowing where someone’s attention is — in remote settings, we lose these nonverbal, implicit cues that are very important for carrying out design activities,” said Mose Sakashita, a doctoral student of information science.
Sakashita is the lead author of “ReMotion: Supporting Remote Collaboration in Open Space with Automatic Robotic Embodiment,” which he presented at the Association for Computing Machinery CHI Conference on Human Factors in Computing Systems in Hamburg, Germany. “With ReMotion, we show that we can enable rapid, dynamic interactions through the help of a mobile, automated robot.”
The lean, nearly six-foot-tall device is outfitted with a monitor for a head, omnidirectional wheels for feet and game-engine software for brains. It automatically mirrors the remote user’s movements — thanks to another Cornell-made device, NeckFace, which the remote user wears to track head and body movements. The motion data is then sent remotely to the ReMotion robot in real-time.
Telepresence robots are not new, but remote users generally need to steer them manually, distracting from the task at hand, researchers said. Other options such as virtual reality and mixed reality collaboration can also require an active role from the user and headsets may limit peripheral awareness, researchers added.
In a small study, nearly all participants reported having a better connection with their remote teammates when using ReMotion compared to an existing telerobotic system. Participants also reported significantly higher shared attention among remote collaborators.
In its current form, ReMotion only works with two users in a one-on-one remote environment, and each user must occupy physical spaces of identical size and layout. In future work, ReMotion developers intend to explore asymmetrical scenarios, like a single remote team member collaborating virtually via ReMotion with multiple teammates in a larger room.
With further development, Sakashita says ReMotion could be deployed in virtual collaborative environments as well as in classrooms and other educational settings.
This research was funded in part by the National Science Foundation and the Nakajima Foundation. More

A team of researchers led by Rensselaer Polytechnic Institute’s Trevor David Rhone, assistant professor in the Department of Physics, Applied Physics, and Astronomy, has identified novel van der Waals (vdW) magnets using cutting-edge tools in artificial intelligence (AI). In particular, the team identified transition metal halide vdW materials with large magnetic moments that are predicted to be chemically stable using semi-supervised learning. These two-dimensional (2D) vdW magnets have potential applications in data storage, spintronics, and even quantum computing.
Rhone specializes in harnessing materials informatics to discover new materials with unexpected properties that advance science and technology. Materials informatics is an emerging field of study at the intersection of AI and materials science. His team’s latest research was recently featured on the cover of Advanced Theory and Simulations.
2D materials, which can be as thin as a single atom, were only discovered in 2004 and have been the subject of great scientific curiosity because of their unexpected properties. 2D magnets are significant because their long-range magnetic ordering persists when they are thinned down to one or a few layers. This is due to magnetic anisotropy. The interplay with this magnetic anisotropy and low dimensionality could give rise to exotic spin degrees of freedom, such as spin textures that can be used in the development of quantum computing architectures. 2D magnets also span the full range of electronic properties and can be used in high-performance and energy-efficient devices.
Rhone and team combined high-throughput density functional theory (DFT) calculations, to determine the vdW materials’ properties, with AI to implement a form of machine learning called semi-supervised learning. Semi-supervised learning uses a combination of labeled and unlabeled data to identify patterns in data and make predictions. Semi-supervised learning mitigates a major challenge in machine learning — the scarcity of labeled data.
“Using AI saves time and money,” said Rhone. “The typical materials discovery process requires expensive simulations on a supercomputer that can take months. Lab experiments can take even longer and can be more expensive. An AI approach has the potential to speed up the materials discovery process.”
Using an initial subset of 700 DFT calculations on a supercomputer, an AI model was trained that could predict the properties of many thousands of materials candidates in milliseconds on a laptop. The team then identified promising candidate vdW materials with large magnetic moments and low formation energy. Low formation energy is an indicator of chemical stability, which is an important requirement for synthesizing the material in a laboratory and subsequent industrial applications.
“Our framework can easily be applied to explore materials with different crystal structures, as well,” said Rhone. “Mixed crystal structure prototypes, such as a data set of both transition metal halides and transition metal trichalcogenides, can also be explored with this framework.”
“Dr. Rhone’s application of AI to the field of materials science continues to produce exciting results,” said Curt Breneman, dean of Rensselaer’s School of Science. “He has not only accelerated our understanding of 2D materials that have novel properties, but his findings and methods are likely to contribute to new quantum computing technologies.”
Rhone was joined in research by Romakanta Bhattarai and Haralambos Gavras of Renselaer; Bethany Lusch and Misha Salim of Argonne National Laboratory; Marios Mattheakis, Daniel T. Larson, and Efthimios Kaxiras of Harvard University; and Yoshiharu Krockenberger of NTT Basic Research Laboratories. More

Privacy and security features that aim to give consumers more control over the sharing of their data by smartphone apps are widely misunderstood, shows new research from the University of Bath’s School of Management.
43 per cent of phone users in the study were confused or unclear about what app tracking means. People commonly mistook the purpose of tracking, thinking that it was intrinsic to the app function, or that it would provide a better user experience.
App tracking is used by companies to deliver targeted advertising to smartphone users.
When iPhone users first open an app, a pop-up asks whether they want to allow the app company to track their activity across other apps. They can choose either ‘Ask App Not to Track’ or ‘Allow’, as introduced by Apple’s App Tracking Transparency framework in April 2021. Android users must access tracking consent via their phone settings.
If people opt out of tracking, their use of apps and websites on their device can no longer be traced by the company, and the data can’t be used for targeted advertising, or shared with data brokers.
The most common misapprehension (24 per cent) was that tracking refers to sharing the physical location of the device — rather than tracing the use of apps and websites. People thought they needed to accept tracking for food delivery and collection services, such as Deliveroo, or for health and fitness apps, because they believed their location was integral to the functioning of the app.

While just over half of participants (51 per cent) said they were concerned about privacy or security — including security of their data after it had been collected — analysis showed no association between their concern for privacy in their daily life and a lower rate of tracking acceptance.
“We asked people about their privacy concerns and expected to see people who are concerned about protecting their privacy allowing fewer apps to track their data, but this wasn’t the case,” said Hannah Hutton, postgraduate researcher from the University of Bath’s School of Management. “There were significant misunderstandings about what app tracking means. People commonly believed they needed to allow tracking for the app to function correctly.
“Some of the confusion is likely to be due to lack of clarity in wording chosen by companies in the tracking prompts, which are easy to misinterpret. For example, when ASOS said ‘We’ll use your data to give you a more personalised ASOS experience and to make our app even more amazing’ it’s probably no surprise that people thought they were opting for additional functionality rather than just more relevant adverts.”
Although the main text of the prompt for app tracking consent is standardised, app developers can include a sentence explaining why they are requesting tracking permission, and this can open the door to false or misleading information, either intentionally or unknowingly.
Other misconceptions included believing that consenting to sharing for health apps (such as period tracking apps) would mean private data being shared, or that denying tracking would remove adverts from the app.
The study, Exploring User Motivations Behind iOS App Tracking Transparency Decisions, is published in the proceedings of The ACM CHI Conference on Human Factors in Computing Systems and was presented at the CHI23 conference in Hamburg, Germany (23-28 April). It isthought to be the first academic analysis of the decisions people make when faced with tracking requests.
The researchers collected data on the tracking decisions of 312 study participants (aged 18 to 75) and analysed reasons for allowing or rejecting tracking across a range of apps, including social media, shopping, health, and food delivery.
David Ellis, a Professor of Behavioural Science and co-author, added: “This research further exposes how most consumers are not aware of how their digital data is being used. Everyday millions of us share information with tech companies and while some of this data is essential for these services to function correctly, other data allows them to generate money from advertising revenue. For example, Meta predicted that they would lose $10Billion from people rejecting tracking.
“While people are now familiar with the benefits of having PIN numbers and facial recognition to protect our devices, more work needs to be done so people can make transparent decisions about what other data is used for in the digital age.” More

Espresso coffee is brewed by first grinding roasted coffee beans into grains. Hot water then forces its way through a bed of coffee grains at high pressure, and the soluble content of the coffee grains dissolves into the water (extraction) to produce espresso.
In 2020, researchers found that more finely ground coffee beans brew a weaker espresso. This counterintuitive experimental result makes sense if, for some reason, regions exist within the coffee bed where less or even no coffee is extracted. This uneven extraction becomes more pronounced when coffee is ground more finely.
In Physics of Fluids, from AIP Publishing, University of Huddersfield researchers explored the role of uneven coffee extraction using a simple mathematical model. They split the coffee into two regions to examine whether uneven flow does in fact make weaker espresso.
One of the regions in the model system hosted more tightly packed coffee than the other, which caused an initial disparity in flow resistance because water flows more quickly through more tightly packed grains. The extraction of coffee decreased the flow resistance further, as coffee grains lose about 20% to 25% of their mass during the process.
“Our model shows that flow and extraction widened the initial disparity in flow between the two regions due to a positive feedback loop, in which more flow leads to more extraction, which in turn reduces resistance and leads to more flow,” said co-author William Lee. “This effect appears to always be active, and it isn’t until one of the regions has all of its soluble coffee extracted that we see the experimentally observed decrease in extraction with decreasing grind size.”
The researchers were surprised to find the model always predicts uneven flow across different parts of the coffee bed.
“This is important because the taste of the coffee depends on the level of extraction,” said Lee. “Too little extraction and the taste of the coffee is what experts call ‘underdeveloped,’ or as I describe it: smoky water. Too much extraction and the coffee tastes very bitter. These results suggest that even if it looks like the overall extraction is at the right level, it might be due to a mixture of underdeveloped and bitter coffee.”
Understanding the origin of uneven extraction and avoiding or preventing it could enable better brews and substantial financial savings by using coffee more efficiently.
“Our next step is to make the model more realistic to see if we can obtain more detailed insights into this confusing phenomenon,” said Lee. “Once this is achieved, we can start to think about whether it is possible to make changes to the way espresso coffee is brewed to reduce the amount of uneven extraction.” More

Imagine a cocktail party full of 3D-printed, humanoid robots listening and talking to each other. That seemingly sci-fi scene is the goal of the Augmented Listening Laboratory at the University of Illinois Urbana-Champaign. Realistic talking (and listening) heads are crucial for investigating how humans receive sound and developing audio technology.
The team will describe the talking human head simulators in their presentation, “3D-printed acoustic head simulators that talk and move,” on May 8. Eastern U.S. in the Northwestern/Ohio State room of the Chicago Marriott Downtown Magnificent Mile Hotel. The talk comes as part of the 184th Meeting of the Acoustical Society of America running May 8-12.
Algorithms used to improve human hearing must consider the acoustic properties of the human head. For example, hearing aids adjust the sound received at each ear to create a more realistic listening experience. For the adjustment to succeed, an algorithm must realistically assess the difference between the arrival time at each ear and amplitude of the sound.
It is important to study human listening in natural environments, like cocktail parties, where many conversations occur at once.
“Simulating realistic scenarios for conversation enhancement often requires hours of recording with human subjects. The entire process can be exhausting for the subjects, and it is extremely hard for a subject to remain perfectly still in between and during recordings, which affects the measured acoustic pressures,” said Austin Lu, a student member of the team. “Acoustic head simulators can overcome both drawbacks. They can be used to create large data sets with continuous recording and are guaranteed to remain still.”
Since researchers have precise control over the simulated subject, they can adjust the parameters of the experiment and even set the machines in motion to simulate neck movements.
In a feat of design and engineering, the heads are 3D-printed into components and assembled, enabling customization at low cost. The highly detailed ears are fitted with microphones along different parts to simulate both human hearing and Bluetooth earpieces. The “talkbox,” or mouthlike loudspeaker, closely mimics human vocals. To facilitate motion, the researchers paid special attention to the neck. Because the 3D model of the head design is open source, other teams can download and modify it as needed. The diminishing cost of 3D printing means there is a relatively low barrier for fabricating these heads.
“Our acoustic head project is the culmination of the work done by many students with highly varied technical backgrounds,” said Manan Mittal, a graduate researcher with the team. “Projects like this are due to interdisciplinary research that requires engineers to work with designers.”
The Augmented Listening Laboratory has also created wheeled and pully-driven systems to simulate walking and more complex motion. More

Much of the human genome is made of regulatory regions that control which genes are expressed at a given time within a cell. Those regulatory elements can be located near a target gene or up to 2 million base pairs away from the target.
To enable those interactions, the genome loops itself in a 3D structure that brings distant regions close together. Using a new technique, MIT researchers have shown that they can map these interactions with 100 times higher resolution than has previously been possible.
“Using this method, we generate the highest-resolution maps of the 3D genome that have ever been generated, and what we see are a lot of interactions between enhancers and promoters that haven’t been seen previously,” says Anders Sejr Hansen, the Underwood-Prescott Career Development Assistant Professor of Biological Engineering at MIT and the senior author of the study. “We are excited to be able to reveal a new layer of 3D structure with our high resolution.”
The researchers’ findings suggest that many genes interact with dozens of different regulatory elements, although further study is needed to determine which of those interactions are the most important to the regulation of a given gene.
“Researchers can now affordably study the interactions between genes and their regulators, opening a world of possibilities not just for us but also for dozens of labs that have already expressed interest in our method,” says Viraat Goel, an MIT graduate student and one of the lead authors of the paper. “We’re excited to bring the research community a tool that help them disentangle the mechanisms driving gene regulation.”
MIT postdoc Miles Huseyin is also a lead author of the paper, which appears today in Nature Genetics.

High-resolution mapping
Scientists estimate that more than half of the genome consists of regulatory elements that control genes, which make up only about 2 percent of the genome. Genome-wide association studies, which link genetic variants with specific diseases, have identified many variants that appear in these regulatory regions. Determining which genes these regulatory elements interact with could help researchers understand how those diseases arise and, potentially, how to treat them.
Discovering those interactions requires mapping which parts of the genome interact with each other when chromosomes are packed into the nucleus. Chromosomes are organized into structural units called nucleosomes — strands of DNA tightly wound around proteins — helping the chromosomes fit within the small confines of the nucleus.
Over a decade ago, a team that included researchers from MIT developed a method called Hi-C, which revealed that the genome is organized as a “fractal globule,” which allows the cell to tightly pack its DNA while avoiding knots. This architecture also allows the DNA to easily unfold and refold when needed.
To perform Hi-C, researchers use restriction enzymes to chop the genome into many small pieces and biochemically link pieces that are near each other in 3D space within the cell’s nucleus. They then determine the identities of the interacting pieces by amplifying and sequencing them.

While Hi-C reveals a great deal about the overall 3D organization of the genome, it has limited resolution to pick out specific interactions between genes and regulatory elements such as enhancers. Enhancers are short sequences of DNA that can help to activate the transcription of a gene by binding to the gene’s promoter — the site where transcription begins.
To achieve the resolution necessary to find these interactions, the MIT team built on a more recent technology called Micro-C, which was invented by researchers at the University of Massachusetts Medical School, led by Stanley Hsieh and Oliver Rando. Micro-C was first applied in budding yeast in 2015 and subsequently applied to mammalian cells in three papers in 2019 and 2020 by researchers including Hansen, Hsieh, Rando and others at University of California at Berkeley and at UMass Medical School.
Micro-C achieves higher resolution than Hi-C by using an enzyme known as micrococcal nuclease to chop up the genome. Hi-C’s restriction enzymes cut the genome only at specific DNA sequences that are randomly distributed, resulting in DNA fragments of varying and larger sizes. By contrast, micrococcal nuclease uniformly cuts the genome into nucleosome-sized fragments, each of which contains 150 to 200 DNA base pairs. This uniformity of small fragments grants Micro-C its superior resolution over Hi-C.
However, since Micro-C surveys the entire genome, this approach still doesn’t achieve high enough resolution to identify the types of interactions the researchers wanted to see. For example, if you want to look at how 100 different genome sites interact with each other, you need to sequence at least 100 multiplied by 100 times, or 10,000. The human genome is very large and contains around 22 million sites at nucleosome resolution. Therefore, Micro-C mapping of the entire human genome would require at least 22 million multiplied by 22 million sequencing reads, costing more than $1 billion.
To bring that cost down, the team devised a way to perform a more targeted sequencing of the genome’s interactions, allowing them to focus on segments of the genome that contain genes of interest. By focusing on regions spanning few million base pairs, the number of possible genomic sites decreases a thousandfold and the sequencing costs decrease a millionfold, down to about $1,000. The new method, called Region Capture Micro-C (RCMC), is therefore able to inexpensively generate maps 100 times richer in information than other published techniques for a fraction of the cost.
“Now we have a method for getting ultra-high-resolution 3D genome structure maps in a very affordable manner. Previously, it was so inaccessible financially because you would need millions, if not billions of dollars, to get high resolution,” Hansen says. “The one limitation is that you can’t get the whole genome, so you need to know approximately what region you’re interested in, but you can get very high resolution, very affordably.”
Many interactions
In this study, the researchers focused on five regions varying in size from hundreds of thousands to about 2 million base pairs, which they chose due to interesting features revealed by previous studies. Those include a well-characterized gene called Sox2, which plays a key role in tissue formation during embryonic development.
After capturing and sequencing the DNA segments of interest, the researchers found many enhancers that interact with Sox2, as well as interactions between nearby genes and enhancers that were previously unseen. In other regions, especially those full of genes and enhancers, some genes interacted with as many as 50 other DNA segments, and on average each interacting site contacted about 25 others.
“People have seen multiple interactions from one bit of DNA before, but it’s usually on the order of two or three, so seeing this many of them was quite significant in terms of difference,” Huseyin says.
However, the researchers’ technique doesn’t reveal whether all of those interactions occur simultaneously or at different times, or which of those interactions are the most important.
The researchers also found that DNA appears to coil itself into nested “microcompartments” that facilitate these interactions, but they weren’t able to determine how microcompartments form. The researchers hope that further study into the underlying mechanisms could shed light on the fundamental question of how genes are regulated.
“Even though we’re not currently aware of what may be causing these microcompartments, and we have all these open questions in front of us, we at least have a tool to really stringently ask those questions,” Goel says.
In addition to pursuing those questions, the MIT team also plans to work with researchers at Boston Children’s Hospital to apply this type of analysis to genomic regions that have been linked with blood disorders in genome-wide association studies. They are also collaborating with researchers at Harvard Medical School to study variants linked to metabolic disorders.
The research was funded by the Koch Institute Support (core) Grant from the National Cancer Institute, the National Institutes of Health, the National Science Foundation, a Solomon Buchsbaum Research Support Committee Award, the Koch Institute Frontier Research Fund, an NIH Fellowship and an EMBO Fellowship. More

Researchers at Columbia Engineering have developed a new class of integrated photonic devices — “leaky-wave metasurfaces” — that can convert light initially confined in an optical waveguide to an arbitrary optical pattern in free space. These devices are the first to demonstrate simultaneous control of all four optical degrees of freedom, namely, amplitude, phase, polarization ellipticity, and polarization orientation — a world record. Because the devices are so thin, transparent, and compatible with photonic integrated circuits (PICs), they can be used to improve optical displays, LIDAR (Light Detection and Ranging), optical communications, and quantum optics.
“We are excited to find an elegant solution for interfacing free-space optics and integrated photonics — these two platforms have traditionally been studied by investigators from different subfields of optics and have led to commercial products addressing completely different needs,” said Nanfang Yu, associate professor of applied physics and applied mathematics who is a leader in research on nanophotonic devices. “Our work points to new ways to create hybrid systems that utilize the best of both worlds — free-space optics for shaping the wavefront of light and integrated photonics for optical data processing — to address many emerging applications such as quantum optics, optogenetics, sensor networks, inter-chip communications, and holographic displays.”
Bridging free-space optics and integrated photonics
The key challenge of interfacing PICs and free-space optics is to transform a simple waveguide mode confined within a waveguide — athin ridge defined on a chip — into a broad free-space wave with a complex wavefront, and vice versa. Yu’s team tackled this challenge by building on their invention last fall of “nonlocal metasurfaces” and extended the devices’ functionality from controlling free-space light waves to controlling guided waves.
Specifically, they expanded the input waveguide mode by using a waveguide taper into a slab waveguide mode — a sheet of light propagating along the chip. “We realized that the slab waveguide mode can be decomposed into two orthogonal standing waves — waves reminiscent of those produced by plucking a string,” said Heqing Huang, a PhD student in Yu’s lab and co-first author of the study, published today in Nature Nanotechnology. “Therefore, we designed a ‘leaky-wave metasurface’ composed of two sets of rectangular apertures that have a subwavelength offset from each other to independently control these two standing waves. The result is that each standing wave is converted into a surface emission with independent amplitude and polarization; together, the two surface emission components merge into a single free-space wave with completely controllable amplitude, phase, and polarization at each point over its wavefront.”
From quantum optics to optical communications to holographic 3D displays
Yu’s team experimentally demonstrated multiple leaky-wave metasurfaces that can convert a waveguide mode propagating along a waveguide with a cross-section on the order of one wavelength into free-space emission with a designer wavefront over an area about 300 times the wavelength at the telecom wavelength of 1.55 microns. These include:

A leaky-wave metalens that produces a focal spot in free space. Such a device will be ideal for forming a low-loss, high-capacity free-space optical link between PIC chips; it will also be useful for an integrated optogenetic probe that produces focused beams to optically stimulate neurons located far away from the probe.
Aleaky-wave optical-lattice generator that can produce hundreds of focal spots forming a Kagome lattice pattern in free space. In general, the leaky-wave metasurface can produce complex aperiodic and three-dimensional optical lattices to trap cold atoms and molecules. This capability will enable researchers to study exotic quantum optical phenomena or conduct quantum simulations hitherto not easily attainable with other platforms, and enable them to substantially reduce the complexity, volume, and cost of atomic-array-based quantum devices. For example, the leaky-wave metasurface could be directly integrated into the vacuum chamber to simplify the optical system, making portable quantum optics applications, such as atomic clocks, a possibility.
A leaky-wave vortex-beam generator that produces a beam with a corkscrew-shaped wavefront. This could lead to a free-space optical link between buildings that relies on PICs to process information carried by light, while also using light waves with shaped wavefronts for high-capacity intercommunication.
A leaky-wave hologram that can displace four distinct images simultaneously: two at the device plane (at two orthogonal polarization states) and another two at a distance in the free space (also at two orthogonal polarization states). This function could be used to make lighter, more comfortable augmented reality goggles and more realistic holographic 3D displays.
Device fabrication
Device fabrication was carried out at the Columbia Nano Initiative cleanroom, and at the Advanced Science Research Center NanoFabrication Facility at the Graduate Center of the City University of New York.

Next steps
Yu’s current demonstration is based on a simple polymer-silicon nitride materials platform at near-infrared wavelengths. His team plans next to demonstrate devices based on the more robust silicon nitride platform, which is compatible with foundry fabrication protocols and tolerant to high optical power operation. They also plan to demonstrate designs for high output efficiency and operation at visible wavelengths, which is more suitable for applications such as quantum optics and holographic displays.
The study was supported by the National Science Foundation (grant no. QII-TAQS-1936359 (H.H., Y.X., and N.Y.) and no. ECCS-2004685 (S.C.M., C.-C.T., and N.Y.)), the Air Force Office of Scientific Research (no. FA9550-16-1-0322 (N.Y.)), and the Simons Foundation (A.C.O. and A.A). S.C.M. acknowledges support from the NSF Graduate Research Fellowship Program (grant no. DGE-1644869). More